Complex Variables

by R. B. Ash and W.P. Novinger Preface

This book represents a substantial revision of the rst edition which was published in 1971. Most of the topics of the original edition have been retained, but in a number of instances the material has been reworked so as to incorporate alternative approaches to these topics that have appeared in the mathematical literature in recent years. The book is intended as a text, appropriate for use by advanced undergraduates or graduate students who have taken a course in introductory real analysis, or as it is often called, advanced calculus. No background in complex variables is assumed, thus making the text suitable for those encountering the subject for the rst time. It should be possible to cover the entire book in two semesters. The list below enumerates many of the major changes and/or additions to the rst edition. 1. The relationship between real-dierentiability and the Cauchy-Riemann equations. 2. J.D. Dixons proof of the homology version of Cauchys theorem. 3. The use of hexagons in tiling the plane, instead of squares, to characterize simple connectedness in terms of winding numbers of cycles. This avoids troublesome details that appear in the proofs where the tiling is done with squares. 4. Sandy Grabiners simplied proof of Runges theorem. 5. A self-contained approach to the problem of extending Riemann maps of the unit disk to the boundary. In particular, no use is made of the Jordan curve theorem, a dicult theorem which we believe to be peripheral to a course in complex analysis. Several applications of the result on extending maps are given. 6. D.J. Newmans proof of the prime number theorem, as modied by J. Korevaar, is presented in the last chapter as a means of collecting and applying many of the ideas and results appearing in earlier chapters, while at the same time providing an introduction to several topics from analytic number theory. For the most part, each section is dependent on the previous ones, and we recommend that the material be covered in the order in which it appears. Problem sets follow most sections, with solutions provided (in a separate section). 1
1

2 We have attempted to provide careful and complete explanations of the material, while at the same time maintaining a writing style which is succinct and to the point. c Copyright 2004 by R.B. Ash and W.P. Novinger. Paper or electronic copies for noncommercial use may be made freely without explicit permission of the authors. All other rights are reserved.

Ch: 1

2 (Preface-2)

TOC

Index

Complex Variables
by Rob ert B. Ash and W.P. Novinger Table Of Contents
Chapter 1: Introduction
1.1 1.2 1.3 1.4 1.5 1.6 Basic Denitions Further Topology of the Plane Analytic Functions Real-Dierentiability and the Cauchy-Riemann Equations The Exponential Function Harmonic Functions

Chapter 2: The Elementary Theory

2.1 2.2 2.3 2.4 Integration on Paths Power Series The Exponential Function and the Complex Trigonometric Functions Further Applications

Chapter 3: The General Cauchy Theorem

3.1 3.2 3.3 3.4 Logarithms and Arguments The Index of a Point with Respect to a Closed Curve Cauchys Theorem Another Version of Cauchys Theorem

Chapter 4: Applications of the Cauchy Theory

4.1 4.2 4.3 4.4 4.5 4.6 Singularities Residue Theory The Open mapping Theorem for Analytic Functions Linear Fractional Transformations Conformal Mapping Analytic Mappings of One Disk to Another 1
Ch: 1 2 3 4 5 6 7 3 (Table of Contents-1) TOC Index

2 4.7 The Poisson Integral formula and its Applications 4.8 The Jensen and Poisson-Jensen Formulas 4.9 Analytic Continuation

Chapter 5: Families of Analytic Functions

5.1 The Spaces A() and C() 5.2 The Riemann Mapping Theorem 5.3 Extending Conformal Maps to the Boundary

Chapter 6: Factorization of Analytic Functions

6.1 Innite Products 6.2 Weierstrass Products 6.3 Mittag-Leers Theorem and Applications

Chapter 7: The Prime Number Theorem

7.1 The Riemann Zeta Function 7.2 An Equivalent Version of the Prime Number Theorem 7.3 Proof of the Prime Number Theorem

Ch: 1

4 (Table of Contents-2)

TOC

Index

Chapter 1

Introduction
The reader is assumed to be familiar with the complex plane C to the extent found in most college algebra texts, and to have had the equivalent of a standard introductory course in real analysis (advanced calculus). Such a course normally includes a discussion of continuity, dierentiation, and Riemann-Stieltjes integration of functions from the real line to itself. In addition, there is usually an introductory study of metric spaces and the associated ideas of open and closed sets, connectedness, convergence, compactness, and continuity of functions from one metric space to another. For the purpose of review and to establish notation, some of these concepts are discussed in the following sections.

1.1

Basic Denitions

The complex plane C is the set of all ordered pairs (a, b) of real numbers, with addition and multiplication dened by (a, b) + (c, d) = (a + c, b + d) and (a, b)(c, d) = (ac bd, ad + bc).

If i = (0, 1) and the real number a is identied with (a, 0), then (a, b) = a + bi. The expression a + bi can be manipulated as if it were an ordinary binomial expression of real numbers, subject to the relation i2 = 1. With the above denitions of addition and multiplication, C is a eld. If z = a + bi, then a is called the real part of z , written a = Re z , and b is called the imaginary part of z , written b = Im z . The absolute value or magnitude or modulus of z is dened as (a2 + b2 )1/2 . A complex number with magnitude 1 is said to be unimodular. An argument of z (written arg z ) is dened as the angle which the line segment from (0, 0) to (a, b) makes with the positive real axis. The argument is not unique, but is determined up to a multiple of 2 . If r is the magnitude of z and is an argument of z , we may write z = r(cos + i sin ) and it follows from trigonometric identities that |z1 z2 | = |z1 ||z2 | and 1
Ch: 1 2 3 4 5 6 7 5 (1-1) TOC Index

arg(z1 z2 ) = arg z1 + arg z2

CHAPTER 1. INTRODUCTION

(that is, if k is an argument of zk , k = 1, 2, then 1 + 2 is an argument of z1 z2 ). If z2 = 0, then arg(z1 /z2 ) = arg(z1 ) arg(z2 ). If z = a + bi, the conjugate of z is dened as z = a bi, and we have the following properties: |z | = |z |, arg z = arg z, z1 + z 2 = z 1 + z 2 , z1 z2 = z 1 z 2 , zz = |z |2 .

z1 z2 = z 1 z 2 ,

Re z = (z + z )/2,

Im z = (z z )/2i,

The distance between two complex numbers z1 and z2 is dened as d(z1 , z2 ) = |z1 z2 |. So d(z1 , z2 ) is simply the Euclidean distance between z1 and z2 regarded as points in the plane. Thus d denes a metric on C, and furthermore, d is complete, that is, every Cauchy sequence converges. If z1 , z2 , . . . is sequence of complex numbers, then zn z if and only if Re zn Re z and Im zn Im z . We say that zn if the sequence of real numbers |zn | approaches +. Many of the above results are illustrated in the following analytical proof of the triangle inequality: |z1 + z2 | |z1 | + |z2 | for all z1 , z2 C. The geometric interpretation is that the length of a side of a triangle cannot exceed the sum of the lengths of the other two sides. See Figure 1.1.1, which illustrates the familiar representation of complex numbers as vectors in the plane. oo  z1 +z2 ooo o o  z2 oo //  77 oo ??

oo

z1

Figure 1.1.1 The proof is as follows: |z1 + z2 |2 = (z1 + z2 )(z 1 + z 2 ) = |z1 |2 + |z2 |2 + z1 z 2 + z 1 z2 = |z1 |2 + |z2 |2 + z1 z 2 + z1 z 2 = |z1 |2 + |z2 |2 + 2 Re(z1 z 2 ) |z1 |2 + |z2 |2 + 2|z1 z 2 | = (|z1 | + |z2 |)2 . The proof is completed by taking the square root of both sides. If a and b are complex numbers, [a, b] denotes the closed line segment with endpoints a and b. If t1 and t2 are arbitrary real numbers with t1 < t2 , then we may write [a, b] = {a + t t1 (b a) : t1 t t2 }. t2 t 1

The notation is extended as follows. If a1 , a2 , . . . , an+1 are points in C, a polygon from a1 to an+1 (or a polygon joining a1 to an+1 ) is dened as
n

[aj , aj +1 ],
j =1

often abbreviated as [a1 , . . . , an+1 ].

Ch: 1

6 (1-2)

TOC

Index

1.2. FURTHER TOPOLOGY OF THE PLANE

1.2

Further Topology of the Plane

Recall that two subsets S1 and S2 of a metric space are separated if there are open sets G1 S1 and G2 S2 such that G1 S2 = G2 S1 = , the empty set. A set is connected i it cannot be written as the union of two nonempty separated sets. An open (respectively closed) set is connected i it is not the union of two nonempty disjoint open (respectively closed) sets.

1.2.1

Denition

A set S C is said to be polygonally connected if each pair of points in S can be joined by a polygon that lies in S . Polygonal connectedness is a special case of path (or arcwise) connectedness, and it follows that a polygonally connected set, in particular a polygon itself, is connected. We will prove in Theorem 1.2.3 that any open connected set is polygonally connected.

1.2.2

Denitions

If a C and r > 0, then D(a, r) is the open disk with center a and radius r; thus D(a, r) = {z : |z a| < r}. The closed disk {z : |z a| r} is denoted by D(a, r), and C (a, r) is the circle with center a and radius r.

1.2.3

Theorem

If is an open subset of C, then is connected i is polygonally connected. Proof. If is connected and a , let 1 be the set of all z in such that there is a polygon in from a to z , and let 2 = \ 1 . If z 1 , there is an open disk D(z, r) (because is open). If w D(z, r), a polygon from a to z can be extended to w, and it follows that D(z, r) 1 , proving that 1 is open. Similarly, 2 is open. (Suppose z 2 , and choose D(z, r) . Then D(z, r) 2 as before.) Thus 1 and 2 are disjoint open sets, and 1 = because a 1 . Since is connected we must have 2 = , so that 1 = . Therefore is polygonally connected. The converse assertion follows because any polygonally connected set is connected.

1.2.4

Denitions

A region in C is an open connected subset of C. A set E C is convex if for each pair of points a, b E , we have [a, b] E ; E is starlike if there is a point a E (called a star center ) such that [a, z ] E for each z E . Note that any nonempty convex set is starlike and that starlike sets are polygonally connected.

Ch: 1

7 (1-3)

TOC

Index

CHAPTER 1. INTRODUCTION

1.3
1.3.1

Analytic Functions
Denition

Let f : C, where is a subset of C. We say that f is complex-dierentiable at the point z0 if for some C we have
h0

lim

f (z0 + h) f (z0 ) = h f (z ) f (z0 ) = . z z0

(1)

or equivalently,
z z0

lim

(2)

Conditions (3), (4) and (5) below are also equivalent to (1), and are sometimes easier to apply.
n

lim

f (z0 + hn ) f (z0 ) = hn

(3)

for each sequence {hn } such that z0 + hn \ {z0 } and hn 0 as n . f (zn ) f (z0 ) = n zn z0 lim for each sequence {zn } such that zn \ {z0 } and zn z0 as n . f (z ) = f (z0 ) + (z z0 )( + (z )) (5) (4)

for all z , where : C is continuous at z0 and (z0 ) = 0. To show that (1) and (5) are equivalent, just note that may be written in terms of f as follows: (z ) =
f (z )f (z0 ) z z0

if z = z0 if z = z0 .

The number is unique. It is usually written as f (z0 ), and is called the derivative of f at z0 . If f is complex-dierentiable at every point of , f is said to be analytic or holomorphic on . Analytic functions are the basic objects of study in complex variables. Analyticity on a nonopen set S C means analyticity on an open set S . In particular, f is analytic at a point z0 i f is analytic on an open set with z0 . If f1 and f2 are analytic on , so are f1 + f2 , f1 f2 , kf1 for k C, f1 f2 , and f1 /f2 (provided that f2 is never 0 on ). Furthermore,
(f1 + f2 ) = f1 + f2 , (f1 f2 ) = f1 f2 , (kf1 ) = kf1

(f1 f2 ) = f1 f2 + f1 f2 ,

f1 f2

f1 f2 f2 f1 . 2 f2

Ch: 1

8 (1-4)

TOC

Index

1.4. REAL-DIFFERENTIABILITY AND THE CAUCHY-RIEMANN EQUATIONS 5 The proofs are identical to the corresponding proofs for functions from R to R. d Since dz (z ) = 1 by direct computation, we may use the rule for dierentiating a product (just as in the real case) to obtain d n (z ) = nz n1 , n = 0, 1, . . . dz This extends to n = 1, 2, . . . using the quotient rule. If f is analytic on and g is analytic on f () = {f (z ) : z }, then the composition g f is analytic on and d g (f (z )) = g (f (z )f (z ) dz just as in the real variable case. As an example of the use of Condition (4) of (1.3.1), we now prove a result that will be useful later in studying certain inverse functions.

1.3.2

Theorem

Let g be analytic on the open set 1 , and let f be a continuous complex-valued function on the open set . Assume (i) f () 1 , (ii) g is never 0, (iii) g (f (z )) = z for all z (thus f is 1-1). Then f is analytic on and f = 1/(g f ). Proof. Let z0 , and let {zn } be a sequence in \ {z0 } with zn z0 . Then f (zn ) f (z0 ) f (zn ) f (z0 ) g (f (zn )) g (f (z0 )) = = zn z0 g (f (zn )) g (f (z0 )) f (zn ) f (z0 )
1

(Note that f (zn ) = f (z0 ) since f is 1-1 and zn = z0 .) By continuity of f at z0 , the expression in brackets approaches g (f (z0 )) as n . Since g (f (z0 )) = 0, the result follows.

1.4

Real-Dierentiability and the Cauchy-Riemann Equations

Let f : C, and set u = Re f, v = Im f . Then u and v are real-valued functions on and f = u + iv . In this section we are interested in the relation between f and its real and imaginary parts u and v . For example, f is continuous at a point z0 i both u and v are continuous at z0 . Relations involving derivatives will be more signicant for us, and for this it is convenient to be able to express the idea of dierentiability of real-valued function of two real variables by means of a single formula, without having to consider partial derivatives separately. We do this by means of a condition analogous to (5) of (1.3.1).

Ch: 1

9 (1-5)

TOC

Index

CHAPTER 1. INTRODUCTION

Convention
From now on, will denote an open subset of C, unless otherwise specied.

1.4.1

Denition

Let g : R. We say that g is real-dierentiable at z0 = x0 + iy0 if there exist real numbers A and B , and real functions 1 and 2 dened on a neighborhood of (x0 , y0 ), such that 1 and 2 are continuous at (x0 , y0 ), 1 (x0 , y0 ) = 2 (x0 , y0 ) = 0, and g (x, y ) = g (x0 , y0 ) + (x x0 )[A + 1 (x, y )] + (y y0 )[B + 2 (x, y )] for all (x, y ) in the above neighborhood of (x0 , y0 ). It follows from the denition that if g is real-dierentiable at (x0 , y0 ), then the partial derivatives of g exist at (x0 , y0 ) and g (x0 , y0 ) = A, x g (x0 , y0 ) = B. y

g g If, on the other hand, x and y exist at (x0 , y0 ) and one of these exists in a neighborhood of (x0 , y0 ) and is continuous at (x0 , y0 ), then g is real-dierentiable at (x0 , y0 ). To verify g is continuous at (x0 , y0 ), and write this, assume that x

g (x, y ) g (x0 , y0 ) = g (x, y ) g (x0 , y ) + g (x0 , y ) g (x0 , y0 ). Now apply the mean value theorem and the denition of partial derivative respectively (Problem 4).

1.4.2

Theorem

Let f : C, u = Re f, v = Im f . Then f is complex-dierentiable at (x0 , y0 ) i u and v are real-dierentiable at (x0 , y0 ) and the Cauchy-Riemann equations u v = , x y v u = x y

are satised at (x0 , y0 ). Furthermore, if z0 = x0 + iy0 , we have f (z0 ) = u v v u (x0 , y0 ) + i (x0 , y0 ) = (x0 , y0 ) i (x0 , y0 ). x x y y

Proof. Assume f complex-dierentiable at z0 , and let be the function supplied by (5) of (1.3.1). Dene 1 (x, , y ) = Re (x, y ), 2 (x, y ) = Im (x, y ). If we take real parts of both sides of the equation f (x) = f (z0 ) + (z z0 )(f (z0 ) + (z )) we obtain u(x, y ) = u(x0 , y0 ) + (x x0 )[Re f (z0 ) + 1 (x, y )] + (y y0 )[ Im f (z0 ) 2 (x, y )]. (1)

Ch: 1

10 (1-6)

TOC

Index

1.5. THE EXPONENTIAL FUNCTION It follows that u is real-dierentiable at (x0 , y0 ) with u (x0 , y0 ) = Re f (z0 ), x u (x0 , y0 ) = Im f (z0 ). y

(2)

Similarly, take imaginary parts of both sides of (1) to obtain v (x, y ) = v (x0 , y0 ) + (x x0 )[Im f (z0 ) + 2 (x, y )] + (y y0 )[Re f (z0 ) + 1 (x, y )] and conclude that v (x0 , y0 ) = Im f (z0 ), x v (x0 , y0 ) = Re f (z0 ). y (3)

The Cauchy-Riemann equations and the desired formulas for f (z0 ) follow from (2) and (3). Conversely, suppose that u and v are real-dierentiable at (x0 , y0 ) and satisfy the Cauchy-Riemann equations there. Then we may write equations of the form u(x, y ) = u(x0 , y0 ) + (x x0 )[ u (x0 , y0 ) + 1 (x, y )] x u + (y y0 )[ (x0 , y0 ) + 2 (x, y )], y v v (x, y ) = v (x0 , y0 ) + (x x0 )[ (x0 , y0 ) + 3 (x, y )] x v + (y y0 )[ (x0 , y0 ) + 4 (x, y )]. y

(4)

(5)

Since f = u + iv , (4) and (5) along with the Cauchy-Riemann equations yield f (z ) = f (z0 ) + (z z0 )[ where, at least in a neighborhood of z0 , (z ) = y y0 x x0 [1 (x, y ) + i3 (x, y )] + [2 (x, y ) + i4 (x, y )] if z = z0 ; (z0 ) = 0. z z0 z z0 u v (x0 , y0 ) + i (x0 , y0 ) + (z )] x x

It follows that f is complex-dierentiable at z0 .

1.5

The Exponential Function

In this section we extend the domain of denition of the exponential function (as normally encountered in calculus) from the real line to the entire complex plane. If we require that the basic rules for manipulating exponentials carry over to the extended function, there is

Ch: 1

11 (1-7)

TOC

Index

CHAPTER 1. INTRODUCTION

only one possible way to dene exp(z ) for z = x + iy C. Consider the following sequence of equations that exp should satisfy: exp(z ) = exp(x + iy ) = exp(x) exp(iy ) = ex 1 + iy + = ex (iy )2 + 2! y2 y4 y3 y5 1 + + i y + 2! 4! 3! 5!

= ex (cos y + i sin y ). Thus we have only one candidate for the role of exp on C.

1.5.1

Denition

If z = x + iy C, let exp(z ) = ex (cos y + i sin y ). Note that if z = x R, then exp(z ) = ex so exp is indeed a extension of the real exponential function.

1.5.2

Theorem
d dz

The exponential function is analytic on C and

exp(z ) = exp(z ) for all z .

Proof. The real and imaginary parts of exp(x + iy ) are, respectively, u(x, y ) = ex cos y and v (x, y ) = ex sin y . At any point (x0 , y0 ), u and v are real-dierentiable (see Problem 4) and satisfy the Cauchy-Riemann equations there. The result follows from (1.4.2). Functions such as exp and the polynomials that are analytic on C are called entire functions. The exponential function is of fundamental importance in mathematics, and the investigation of its properties will be continued in Section 2.3.

1.6
1.6.1

Harmonic Functions
Denition

A function g : R is said to be harmonic on if g has continuous rst and second order partial derivatives on and satises Laplaces equation 2g 2g + =0 x2 y 2 on all of . After some additional properties of analytic functions have been developed, we will be able to prove that the real and imaginary parts of an analytic function on are harmonic on . The following theorem is a partial converse to that result, namely that a harmonic on is locally the real part of an analytic function.

Ch: 1

12 (1-8)

TOC

Index

1.6. HARMONIC FUNCTIONS

1.6.2

Theorem

Suppose u : R is harmonic on , and D is any open disk contained in . Then there exists a function v : D R such that u + iv is analytic on D. The function v is called a harmonic conjugate of u. u Proof. Consider the dierential P dx + Qdy , where P = u y , Q = x . Since u is
Q harmonic, P and Q have continuous partial derivatives on and P y = x . It follows (from calculus) that P dx + Qdy is a locally exact dierential. In other words, there is a function v : D R such that dv = P dx + Qdy . But this just means that on D we have

v u =P = x y

and

v u =Q= . y x

Hence by (1.4.2) (and Problem 4), u + iv is analytic on D.

Problems
1. Prove the parallelogram law |z1 + z2 |2 + |z1 z2 |2 = 2[|z1 |2 + |z2 |2 ] and give a geometric interpretation. 2. Show that |z1 + z2 | = |z1 | + |z2 | i z1 and z2 lie on a common ray from 0 i one of z1 or z2 is a nonnegative multiple of the other. 3. Let z1 and z2 be nonzero complex numbers, and let , 0 , be the angle between them. Show that (a) Re z1 z 2 = |z1 ||z2 | cos , Im z1 z 2 = |z1 ||z2 | sin , and consequently (b) The area of the triangle formed by z1 , z2 and z2 z1 is | Im z1 z 2 |/2.
g g and y exist at (x0 , y0 ) , and suppose that one 4. Let g : R be such that x of these partials exists in a neighborhood of (x0 , y0 ) and is continuous at (x0 , y0 ). Show that g is real-dierentiable at (x0 , y0 ).

5. Let f (x) = z, z C. Show that although f is continuous everywhere, it is nowhere dierentiable. 6. Let f (z ) = |z |2 , z C. Show that f is complex-dierentiable at z = 0, but nowhere else. 7. Let u(x, y ) = |xy |, (x, y ) C. Show that not real-dierentiable at (0,0).
u x

and

u y

both exist at (0,0), but u is

8. Show that the eld of complex numbers is isomorphic to the set of matrices of the form a b with a, b R. 9. Show that the complex eld cannot be ordered. That is, there is no subset P C of positive elements such that (a) P is closed under addition and multiplication. (b) If z P , then exactly one of the relations z P, z = 0, z P holds. b a

Ch: 1

13 (1-9)

TOC

Index

10

CHAPTER 1. INTRODUCTION

10. (A characterization of absolute value) Show that there is a unique function : C R such that (i) (x) = x for all real x 0; (ii) (zw) = (z )(w), z, w C; (iii) is bounded on the unit circle C (0, 1). Hint: First show that (z ) = 1 for |z | = 1. 11. (Another characterization of absolute value) Show that there is a unique function : C R such that (i) (x) = x for all real x 0; (ii) (zw) = (z )(w), z, w C; (iii) (z + w) (z ) + (w), z, w C. 12. Let be a complex number with || < 1. Prove that z = 1 i |z | = 1. 1 z 13. Suppose z C, z = 0. Show that z +
1 z

is real i Im z = 0 or |z | = 1.

14. In each case show that u is harmonic and nd the harmonic conjugate v such that v (0, 0) = 0. (i) u(x, y ) = ey cos x; (ii) u(x, y ) = 2x x3 + 3xy 2 . 15. Let a, b C with a = 0, and let T (z ) = az + b, z C. (i) Show that T maps the circle C (z0 , r) onto the circle C (T (z0 ), r|a|). (ii) For which choices of a and b will T map C (0, 1) onto C (1 + i, 2)? (iii) In (ii), is it possible to choose a and b so that T (1) = 1 + 3i? 16. Show that f (z ) = eRe z is nowhere complex-dierentiable. 17. Let f be a complex-valued function dened on an open set that is symmetric with respect to the real line, that is, z implies z . (Examples are C and D(x, r) where x R.) Set g (z ) = f (z ), and show that g is analytic on if and only if f is analytic on . 18. Show that an equation for the circle C (z0 , r) is zz z 0 z z0 z + z0 z 0 = r2 . 19. (Enestroms theorem) Suppose that P (z ) = a0 + a1 z + + an z n , where n 1 and a0 a1 a2 an > 0. Prove that the zeros of the polynomial P (z ) all lie outside the open unit disk D(0, 1). Hint: Look at (1 z )P (z ), and show that (1 z )P (z ) = 0 implies that a0 = (a0 a1 )z + (a1 a2 )z 2 + + (an1 an )z n + an z n+1 , which is impossible for |z | < 1. 20. Continuing Problem 19, show that if aj 1 > aj for all j , then all the zeros of P (z ) must be outside the closed unit disk D(0, 1). Hint: If the last equation of Problem 19 holds for some z with |z | 1, then z = 1.

Ch: 1

14 (1-10)

TOC

Index

Chapter 2

The Elementary Theory

2.1 Integration on Paths

The integral of a complex-valued function on a path in the complex plane will be introduced via the integral of a complex-valued function of a real variable, which in turn is expressed in terms of an ordinary Riemann integral.

2.1.1

Denition

Let : [a, b] C be a piecewise continuous function on the closed interval [a, b] of reals. The Riemann integral of is dened in terms of the real and imaginary parts of by
b b b

(t) dt =
a a

Re (t) dt + i
a

Im (t) dt.

2.1.2

Basic Properties of the Integral

The following linearity property is immediate from the above denition and the corresponding result for real-valued functions:
b b b

((t) + (t)) dt =
a a

(t) dt +
a

(t) dt

for any complex numbers and . A slightly more subtle property is

b b

(t) dt
a a

|(t)| dt.

This may be proved by approximating the integral on the left by Riemann sums and using the triangle inequality. A somewhat more elegant argument uses a technique called b b polarization, which occurs quite frequently in analysis. Dene = a (t)d t / a (t) dt; then || = 1. (If the denominator is zero, take to be any complex number of absolute 1
Ch: 1 2 3 4 5 6 7 15 (2-1) TOC Index

2
b b

CHAPTER 2. THE ELEMENTARY THEORY

b a

value 1.) Then | a (t) dt| = a (t) dt = value of a complex number is real,
b b

(t)dt by linearity. Since the absolute

b

(t) dt = Re
a a

(t) dt =
a

Re (t) dt

by denition of the integral. But Re |z | |z |, so

b b b

Re (t) dt
a a

|(t)| dt =
a

|(t)| dt

because || = 1. The fundamental theorem of calculus carries over to complex-valued functions. Explicitly, if has a continuous derivative on [a, b], then
x

(x) = (a) +
a

(t) dt
x

for a x b. If is continuous on [a, b] and F (x) = a (t) dt, a x b, then F (x) = (x) for all x in [a, b]. These assertions are proved directly from the corresponding results for real-valued functions.

2.1.3

Denition

A curve in C is a continuous mapping of a closed interval [a, b] into C. If in addition, is piecewise continuously dierentiable, then is called a path. A curve (or path) with (a) = (b) is called a closed curve (or path). The range (or image or trace) of will be denoted by . If is contained in a set S , is said to be a curve (or path ) in S . Intuitively, if z = (t) and t changes by a small amount dt, then z changes by dz = (t) dt. This motivates the denition of the length L of a path :
b

L=
a

| (t)| dt

and also motivates the following denition of the path integral

f (z )dz .

2.1.4

Denition

Let : [a, b] C be a path, and let f be continuous on , that is, f : C is continuous. We dene the integral of f on (or along) by
b

f (z ) dz =
a

f ( (t)) (t) dt.

It is convenient to dene f (z ) dz with replaced by certain point sets in the plane. Specically, if [z1 , z2 ] is a line segment in C, we dene f (z ) dz =
[z1 ,z2 ]

f (z ) dz

Ch: 1

16 (2-2)

TOC

Index

2.1. INTEGRATION ON PATHS

where (t) = (1 t)z1 + tz2 , 0 t 1. More generally, if [z1 , . . . , zn+1 ] is a polygon joining z1 to zn+1 , we dene
n

f (z ) dz =
[z1 ,z2 ,... ,zn+1 ] j =1 [zj ,zj +1 ]

f (z ) dz.

The next estimate will be referred to as the M-L theorem.

2.1.5

Theorem

Suppose that f is continuous on the path and |f (z )| M for all z . If L is the length of the path , then f (z ) dz M L.

Proof. Recall from (2.1.2) that the absolute value of an integral is less than or equal to the integral of the absolute value. Then apply the denition of the path integral in (2.1.4) and the denition of length in (2.1.3). The familiar process of evaluating integrals by anti-dierentiation extends to integration on paths.

2.1.6

Fundamental Theorem for Integrals on Paths

Suppose f : C is continuous and f has a primitive F on , that is, F = f on . Then for any path : [a, b] we have f (z ) dz = F ( (b)) F ( (a)).

In particular, if is a closed path in , then Proof. f (z )dz = F ( (t)) (t) dt = fundamental theorem of calculus [see (2.1.2)].
b a

f (z ) dz = 0. = F ( (b)) F ( (a)) by the

b d F ( (t)) dt a dt

2.1.7

Applications

(a) Let z1 , z2 C and let be any path from z1 to z2 , that is, : [a, b] C is any path such that (a) = z1 and (b) = z2 . Then for n = 0, 1, 2, 3, . . . we have
n+1 n+1 z n dz = (z2 z1 )/(n + 1).

This follows from (2.1.6) and the fact that z n+1 /(n +1) is a primitive of z n . The preceding remains true for n = 2, 3, 4, . . . provided that 0 / and the proof is the same: z n+1 /(n + 1) is a primitive for z n on C \ {0}. But if n = 1, then the conclusion may

Ch: 1

17 (2-3)

TOC

Index

CHAPTER 2. THE ELEMENTARY THEORY

fail as the following important computation shows. Take (t) = eit , 0 t 2 (the unit circle, traversed once in the positive sense). Then 1 dz = z
2 0

ieit dt = 2i = 0. eit

This also shows that f (z ) = 1/z , although analytic on C \ {0}, does not have a primitive on C \ {0}. (b) Suppose f is analytic on the open connected set and f (z ) = 0 for all z . Then f is constant on . Proof. Let z1 , z2 . Since is polygonally connected, there is a (polygonal) path : [a, b] such that (a) = z1 and (b) = z2 . by (2.1.6), f (z ) dz = f (z2 ) f (z1 ). But the left side is zero by hypothesis, and the result follows.

Remark
If we do not assume that is connected, we can prove only that f restricted to any component of is constant. Suppose that a continuous function f on is given. Theorem 2.1.6 and the applications following it suggest that we should attempt to nd conditions on f and/or that are sucient to guarantee that f has a primitive. Let us attempt to imitate the procedure used in calculus when f is a real-valued continuous function on an open interval in R. We begin by assuming is starlike with star center z0 , say. Dene F on by F (z ) =
[z0 ,z ]

f (w) dw.

If z1 , let us try to show that F (z1 ) = f (z1 ). If z is near but unequal to z1 , we have F (z ) F (z1 ) 1 = z z1 z z1 f (w) dw
[z0 ,z ] [z0 ,z1 ]

f (w) dw

and we would like to say, as in the real variables case, that f (w) dw
[z0 ,z ] [z0 ,z1 ]

f (w) dw =
[z1 ,z ]

f (w) dw,

(1)

from which it would follow quickly that (F (z ) F (z1 ))/(z z1 ) f (z1 ) as z z1 . Now if T is the triangle [z0 , z1 , z, z0 ], equation (1) is equivalent to the statement that f (w) dw = 0, but as the example at the end of (2.1.7(a)) suggests, this need not be T true, even for analytic functions f . However, in the present setting, we can make the is the union of T and its interior (the convex hull of T ), then key observation that if T , and in this case, it turns out T . If f is analytic on , it must be analytic on T that T f (w) dw does equal 0. This is the content of Theorem 2.1.8; a somewhat dierent version of this result was rst proved by Augustin-Louis Cauchy in 1825.

Ch: 1

18 (2-4)

TOC

Index

2.1. INTEGRATION ON PATHS

z3 c z1 a
Figure 2.1.1

z2

2.1.8

Cauchys Theorem for Triangles

. Suppose that f is analytic on and T = [z1 , z2 , z3 , z1 ] is any triangle such that T Then T f (z ) dz = 0. Proof. Let a, b, c be the midpoints of [z1 , z2 ], [z2 , z3 ] and [z3 , z1 ] respectively. Consider the triangles [z1 , a, c, z1 ], [z2 , b, a, z2 ], [z3 , c, b, z3 ] and [a, b, c, a] (see Figure 2.1.1). Now the integral of f on T is the sum of the integrals on the four triangles, and it follows from the triangle inequality that if T1 is one of these four triangles chosen so that | T1 f (z ) dz | is as large as possible, then |
T

f (z ) dz | 4|
T1

f (z ) dz |.

Also, if L measures length, then L(T1 ) = 1 2 L(T ), because a line joining two midpoints of a triangle is half as long as the opposite side. Proceeding inductively, we obtain a sequence n+1 T n , and {Tn : n = 1, 2, . . . } of triangles such that L(Tn ) = 2n L(T ), T |
T

f (z ) dz | 4n |
Tn

f (z ) dz |.

(1)

n form a decreasing sequence of nonempty closed and bounded (hence compact) Now the T sets in C whose diameters approach 0 as n . Thus there is a point z0 n=1 Tn . (If i s would have the intersection is empty, then by compactness, some nite collection of T empty intersection.) Since f is analytic at z0 , there is a continuous function : C with (z0 ) = 0 [see (5) of (1.3.1)] and such that f (z ) = f (z0 ) + (z z0 )[f (z0 ) + (z )], z . By (2) and (2.1.7a), we have f (z ) dz =
Tn Tn

(2)

(z z0 )(z ) dz, n = 1, 2, 3, . . . .

(3)

Ch: 1

19 (2-5)

TOC

Index

6 But by the M-L theorem (2.1.5), |

Tn

CHAPTER 2. THE ELEMENTARY THEORY

(z z0 )(z ) dz | sup [|(z )| |z z0 |]L(Tn )

z Tn

n sup |(z )|(L(Tn ))2 since z T

z Tn

sup |(z )|4n (L(T ))2

z Tn

0 as n . Thus by (1) and (3), |

T

f (z ) dz | sup |(z )|(L(T ))2 0

z Tn

as n , because (z0 ) = 0. We conclude that

f (z ) dz = 0.

We may now state formally the result developed in the discussion preceding Cauchys theorem.

2.1.9

Cauchys Theorem for Starlike Regions

Let f be analytic on the starlike region . Then f has a primitive on , and consequently, by (2.1.6), f (z ) dz = 0 for every closed path in . Proof. Let z0 be a star center for , and dene F on by F (z ) = [z0 ,z] f (w) dw. It follows from (2.1.8) and discussion preceding it that F is a primitive for f . We may also prove the following converse to Theorem (2.1.6).

2.1.10

Theorem

If f : C is continuous and primitive on .

f (z ) dz = 0 for every closed path in , then f has a

Proof. We may assume that is connected (if not we can construct a primitive of f on each component of , and take the union of these to obtain a primitive of f on ). So x z0 , and for each z , let z be a polygonal path in from z0 to z . Now dene F on by F (z ) = z f (w) dw, z . Then the discussion preceding (2.1.8) may be repeated without essential change to show that F = f on . (In Equation (1) in that discussion, [z0 , z ] and [z0 , z1 ] are replaced by the polygonal paths z and z1 , but the line segment [z1 , z ] can be retained for all z suciently close to z1 .)

2.1.11

Remarks

(a) If : [a, b] C is a path, we may traverse backwards by considering the path dened by (t) = (a + b t), a t b. Then = and for every continuous function

Ch: 1

20 (2-6)

TOC

Index

2.1. INTEGRATION ON PATHS

1
Figure 2.1.2

f : C, it follows from the denition of the integral and a brief change of variable argument that f (z ) dz =

f (z ) dz.

(b) Similarly, if 1 : [a, b] C and 2 : [c, d] C are paths with 1 (b) = 2 (c), we may attach 2 to 1 via the path (t) = 1 ((1 2t)a + 2tb), 0 t 1/2 2 ((2 2t)c + (2t 1)d), 1/2 t 1.

Then = 1 2 and for every continuous function f : C,

f (z ) dz =
1

f (z ) dz +
2

f (z ) dz.

There is a technical point that should be mentioned. The path 1 (t), a t b, is strictly speaking not the same as the path 1 ((1 2t)a + 2tb), 0 t 1/2, since they have dierent domains of denition. Given the path 1 : [a, b] C, we are forming a new path and for = 1 h, where h(t) = (1 2t)a + 2tb, 0 t 1/2. It is true then that = 1 every continuous function f on 1 , 1 f (z ) dz = f (z ) dz . Problem 4 is a general result of this type. (c) If 1 and 2 are paths with the same initial point and the same terminal point, we may form a closed path by rst traversing 1 and then traversing 2 backwards. If f is continuous on , then f (z ) dz = 0 i 1 f (z ) dz = 2 f (z ) dz (see Figure 2.1.2).

An Application of 2.1.9
Let = [z1 , z2 , z3 , z4 , z1 ] be a rectangle with center at 0 (see Figure 2.1.3); let us calculate 1 dz . Let be a circle that circumscribes the rectangle , and let 1 , 2 , 3 , 4 be the z arcs of joining z1 to z2 , z2 to z3 , z3 to z4 and z4 to z1 respectively. There is an open half plane (a starlike region) excluding 0 but containing both [z1 , z2 ] and 1 . By (2.1.9) and Remark (2.1.11c), the integral of 1/z on [z1 , z2 ] equals the integral of 1/z on 1 . By considering the other segments of and the corresponding arcs of , we obtain 1 dz = z 1 dz = 2i z

Ch: 1

21 (2-7)

TOC

Index

CHAPTER 2. THE ELEMENTARY THEORY

z 0

z2

Figure 2.1.3

z3

z b a z1
Figure 2.1.4 by a direct calculation, as in (2.1.7a).

The reader who feels that the machinery used to obtain such a simple result is excessive 1 is urged to attempt to compute z dz directly. The following strengthened form of Cauchys Theorem for triangles and for starlike regions will be useful in the next section.

2.1.12

Extended Cauchy Theorem for Triangles

, Let f be continuous on and analytic on \ {z0 }. If T is any triangle such that T then T f (z ) dz = 0. , the result follows from (2.1.8), Cauchys theorem Proof. Let T = [z1 , z2 , z3 , z1 ]. If z0 /T for triangles. Also, if z1 , z2 and z3 are collinear, then T f (z ) dz = 0 for any continuous (not necessarily analytic) function. Thus assume that z1 , z2 and z3 are non-collinear and . Suppose rst that z0 is a vertex, say z0 = z1 . Choose points a [z1 , z2 ] and that z0 T b [z1 , z3 ]; see Figure 2.1.4. By (2.1.9), f (z ) dz =
T [z1 ,a]

f (z ) dz +
[a,b]

f (z ) dz +
[b,z1 ]

f (z ). dz

Ch: 1

22 (2-8)

TOC

Index

2.1. INTEGRATION ON PATHS

z3

z2

Figure 2.1.5 Since f is continuous at z0 = z1 , each of the integrals on the right approaches zero as a, b z1 , by the M-L theorem. Therefore T f (z ) dz = 0. is not a vertex, join z0 to each vertex of T by straight line segments (see If z0 T Figure 2.1.5), and write T f (z ) dz as a sum of integrals, each of which is zero by the above argument.

2.1.13

Extended Cauchy Theorem for Starlike Regions

Let f be continuous on the starlike region and analytic on \ {z0 }. Then f has a primitive on , and consequently f (z ) dz = 0 for every closed path in . Proof. Exactly as in (2.1.9), using (2.1.12) instead of (2.1.8).

Problems
1. Evaluate 2. Evaluate 3. Evaluate
[i,1+2i]

Im z dz .

z dz where traces the arc of the parabola y = x2 from (1,1) to (2,4). f (z ) dz where z1 = i, z2 = 2 + 5i, z3 = 5i and f (x + iy ) = x2 + iy .

[z1 ,z2 ,z3 ]

4. Show that f (z ) dz is independent of the parametrization of in the following sense. Let h : [c, d] [a, b] be one-to-one and continuously dierentiable, with h(c) = a and h(d) = b ( is assumed to be dened on [a, b]). Let 1 = h. Show that 1 is a path, and prove that if f is continuous on , then 1 f (z ) dz = f (z ) dz . 5. In the next section it will be shown that if f is analytic on , then f is also analytic, in particular continuous, on . Anticipating this result, we can use (2.1.6), the fundamental theorem for integration along paths, to show that f (z ) dz = f ( (b)) f ( (a)). Prove the following. (a) If is convex and Re f > 0 on , then f is one-to-one. (Hint: z1 , z2 with z1 = z2 implies that Re[(f (z2 ) f (z1 ))/(z2 z1 )] > 0.) (b) Show that (a) does not generalize to starlike regions. (Consider z + 1/z on a suitable region.)

Ch: 1

23 (2-9)

TOC

Index

10

CHAPTER 2. THE ELEMENTARY THEORY (c) Suppose z0 and f (z0 ) = 0. Show that there exists r > 0 such that f is oneto-one on D(z0 , r). Consequently, if f has no zeros in , then f is locally one-to-one.

2.2

Power Series

In this section we develop the basic facts about complex series, especially complex power series. The main result is that f is analytic at z0 i f can be represented as a convergent power series throughout some neighborhood of z0 . We rst recall some elementary facts about complex series in general.

2.2.1

Denition

Given a sequence w0 , w1 , w2 , . . . of complex numbers, consider the series n=0 wn . If n limn k=0 wk exists and is the complex number w, we say that the series converges to w and write w = n=0 wn . Otherwise, the series is said to diverge. n A useful observation is that a series is convergent i the partial sums k=0 wk form n a Cauchy sequence, that is, k=m wk 0 as m, n . The series n=0 wn is said to converge absolutely if the series n=0 |wn | is convergent. As in the real variables case, an absolutely convergent series is convergent. A necessary and sucient condition for absolute convergence is that the sequence of partial sums n k=0 |wk | be bounded. The two most useful tests for absolute convergence of complex series are the ratio and root tests.

2.2.2

The Ratio Test

n+1 If wn is a series of nonzero terms and if lim supn | ww | < 1, then the series converges n wn+1 absolutely. If lim inf n | wn | > 1, the series diverges.

2.2.3

The Root Test

Let wn be any complex series. If lim supn |wn |1/n < 1, the series converges absolutely, while if lim supn |wn |1/n > 1, the series diverges. The ratio test is usually (but not always) easier to apply in explicit examples, but the root test has a somewhat wider range of applicability and, in fact, is the test that we are going to use to obtain some basic properties of power series. Proofs and a discussion of the relative utility of the tests can be found in most texts on real analysis. We now consider sequences and series of complex-valued functions.

2.2.4

Theorem

Let {fn } be sequence of complex-valued functions on a set S . Then {fn } converges pointwise on S (that is, for each z S , the sequence {fn (z )} is convergent in C) i {fn } is pointwise Cauchy (that is, for each z S , the sequence {fn (z )} is a Cauchy sequence in C). Also, {fn } converges uniformly i {fn } is uniformly Cauchy on S , in other words, |fn (z ) fm (z )| 0 as m, n , uniformly for z S .

Ch: 1

24 (2-10)

TOC

Index

2.2. POWER SERIES

11

(The above result holds just as well if the fn take their values in an arbitrary complete metric space.) Proof. As in the real variables case; see Problem 2.2.1. The next result gives the most useful test for uniform convergence of innite series of functions.

2.2.5

The Weierstrass M -Test

Let g1 , g2 , . . . be complex-valued functions on a set S , and assume that |gn (z )| Mn for all z S . If n=1 Mn < +, then the series n=1 gn (z ) converges uniformly on S . n Proof. Let fn = k=1 gk ; it follows from the given hypothesis that {fn } is uniformly Cauchy on S . The result now follows from (2.2.4). We now consider power series, which are series of the form n=0 an (z z0 )n , where z0 and the an are complex numbers. Thus we are dealing with series of functions n=0 fn n of a very special type, namely fn (z ) = an (z z0 ) . Our rst task is to describe the sets S C on which such a series will converge.

2.2.6
n=0

Theorem

If an (z z0 )n converges at the point z with |z z0 | = r, then the series converges absolutely on D(z0 , r), uniformly on each closed subdisk of D(z0 , r), hence uniformly on each compact subset of D(z0 , r). Proof. We have |an (z z0 )n | = |an (z z0 )n | an (z z0 )n 0, hence the sequence {an (z
n z z0 . The convergence at z implies that z z0 n z0 ) } is bounded. If |z z0 | r < r, then

r z z0 <1 z z0 r proving absolute convergence at z (by comparison with a geometric series). The Weierstrass M -test shows that the series converges uniformly on D(z0 , r ). We now describe convergence in terms of the coecients an .

2.2.7

Theorem

Let n=0 an (z z0 )n be a power series. Let r = [lim supn (|an |1/n )]1 , the radius of convergence of the series. (Adopt the convention that 1/0 = , 1/ = 0.) The series converges absolutely on D(z0 , r), uniformly on compact subsets. The series diverges for |z z0 | > r. Proof. We have lim supn |an (z z0 )n |1/n = (|z z0 |)/r, which will be less than 1 if |z z0 | < r. By (2.2.3), the series converges absolutely on D(z0 , r). Uniform convergence on compact subsets follows from (2.2.6). (We do not necessarily have convergence for |z z0 | = r, but we do have convergence for |z z0 | = r , where r < r can be chosen arbitrarily close to r.) If the series converges at some point z with |z z0 | > r, then by (2.2.6) it converges absolutely at points z such that r < |z z0 | < |z z0 |. But then (|z z0 |)/r > 1, contradicting (2.2.3).

Ch: 1

25 (2-11)

TOC

Index

12

CHAPTER 2. THE ELEMENTARY THEORY

2.2.8

Denition

Let C (z0 , r) denote the circle with center z0 and radius r. then C (z0 ,r) f (z ) dz is dened as f (z ) dz where (t) = z0 + reit , 0 t 2 . The following result provides the essential equipment needed for the theory of power series. In addition, it illustrates the striking dierence between the concept of dierentiability of complex functions and the analogous idea in the real case. We are going to show that if f is analytic on a closed disk, then the value of f at any interior point is completely determined by its values on the boundary, and furthermore there is an explicit formula describing the dependence.

2.2.9

Cauchys Integral Formula for a Circle

1 2i f (w) dw, wz

Let f be analytic on and let D(z0 , r) be a disk such that D(z0 , r) . Then f (z ) = z D(z0 , r).

C (z0 ,r )

Proof. Let D(z0 , ) be a disk such that D(z0 , r) D(z0 , ) . Fix z D(z0 , r) and dene a function g on D(z0 , ) by g (w) =
f (w)f (z ) wz

f (z )

if w = z if w = z.

Then g is continuous on D(z0 , ) and analytic on D(z0 , ) \ {z }, so we may apply (2.1.13) to get C (z0 ,r) g (w) dw = 0. Therefore 1 2i Now 1 dw = wz 1 dw = (w z0 ) (z z0 ) (z z0 )n dw n+1 C (z0 ,r ) n=0 (w z0 )

C (z0 ,r )

f (w) f (z ) dw = wz 2i

C (z0 ,r )

1 dw. wz

C (z0 ,r )

C (z0 ,r )

The series converges uniformly on C (z0 , r) by the Weierstrass M -test, and hence we may integrate term by term to obtain

(z z0 )n
n=0 C (z0 ,r )

1 dw. (w z0 )n+1

But on C (z0 , r) we have w = z0 + reit , 0 t 2 , so the integral on the right is, by (2.2.8),
2

r(n+1) ei(n+1)t ireit dt =

0

0 if n = 1, 2, . . . 2i if n = 0.

Ch: 1

26 (2-12)

TOC

Index

2.2. POWER SERIES We conclude that

1 C (z0 ,r ) wz

13 dw = 2i, and the result follows.

The integral appearing in Cauchys formula is an example of what is known as an integral of the Cauchy type. The next result, which will be useful later, deals with these integrals.

2.2.10

Theorem

Let be a path (not necessarily closed) and let g be a complex-valued continuous function on . Dene a function F on the open set = C \ by F (z ) =

g (w) dw. wz

Then F has derivatives of all orders on , and F (n) (z ) = n!

g (w) dw (w z )n+1

for all z and all n = 0, 1, 2, . . . (take F (0) = F ). Furthermore, F (n) (z ) 0 as |z | . Proof. We use an induction argument. The formula for F (n) (z ) is valid for n = 0, by hypothesis. Assume that the formula holds for a given n and all z ; x z1 and choose r > 0 small enough that D(z1 , r) . For any point z D(z1 , r) with z = z1 we have F (n) (z ) F (n) (z1 ) (n + 1)! z z1 n! z z1 n! z z1 g (w) dw (w z1 )n+2 g (w) (w z1 )n+2 g (w) dw (w z1 )n+2

(w z1 )n+1 (w z )n+1 g (w) dw (n + 1)! (w z )n+1 (w z1 )n+1

n

(1)

(z z1 ) k=0 (w z1 )nk (w z )k g (w) dw (n + 1)! (w z )n+1 (w z1 )n+1

(2)

where the numerator of the rst integral in (2) is obtained from that in (1) by applying n the algebraic identity an+1 bn+1 = (a b) k=0 ank bk with a = w z1 and b = w z . Thus F (n) (z ) F (n) (z1 ) (n + 1)! z z1
n k=0 (w

g (w) dw (w z1 )n+2

= n!

z1 )nk+1 (w z )k (n + 1)(w z )n+1 g (w) dw (w z )n+1 (w z1 )n+2

Ch: 1

27 (2-13)

TOC

Index

14 n! max
w n k=0 (w

CHAPTER 2. THE ELEMENTARY THEORY z1 )nk+1 (w z )k (n + 1)(w z )n+1 g (w) (w z )n+1 (w z1 )n+2

L( )

by the M-L theorem. But the max that appears in brackets approaches 0 as z z1 , since n n nk+1 (w z )k k=0 (w z1 )n+1 = (n + 1)(w z1 )n+1 . Hence k=0 (w z1 ) F (n) (z ) F (n) (z1 ) (n + 1)! z z1 g (w) dw (w z1 )n+2

as z z1 , and the statement of the theorem follows by induction. The fact that |F (n) (z )| 0 as |z | is a consequence of the M-L theorem; specically, F (n) (z )| n! max
w

|g (w)| L( ). |w z |n+1

Theorems 2.2.9 and 2.2.10 now yield some useful corollaries.

2.2.11

Corollary

If f is analytic on , then f has derivatives of all orders on . Moreover, if D(z0 , r) , then f (n) (z ) = n! 2i
C (z0 ,r )

f (w) dw, (w z )n+1

z D(z0 , r).

Proof. Apply (2.2.10) to the Cauchy integral formula (2.2.9).

2.2.12

Corollary

If f has a primitive on , then f is analytic on . Proof. Apply (2.2.11) to any primitive for f .

2.2.13

Corollary

If f is continuous on and analytic on \ {z0 }, then f is analytic on . Proof. Choose any disk D such that D . By (2.1.13), f has a primitive on D, hence by (2.2.12), f is analytic on D. It follows that f is analytic on . The next result is a converse to Cauchys theorem for triangles.

2.2.14

Moreras Theorem

. Suppose f is continuous on and T f (z ) dz = 0 for each triangle T such that T Then f is analytic on . Proof. Let D be any disk contained in . The hypothesis implies that f has a primitive on D [see the discussion preceding (2.1.8)]. Thus by (2.2.12), f is analytic on D. Since D is an arbitrary disk in , f is analytic on . One of many applications of Moreras theorem is the Schwarz reection principle, which deals with the problem of extending an analytic function to a larger domain.

Ch: 1

28 (2-14)

TOC

Index

2.2. POWER SERIES

15

2.2.15

The Schwarz Reection Principle

Then f is analytic on C \ R and continuous on C (Problem 10). One can then use Moreras theorem to show that f is analytic on C (Problem 11). We now complete the discussion of the connection between analytic functions and power series, showing in essence that the two notions are equivalent. We say that a function f : C is representable in by power series if given D(z0 , r) , there is a sequence {an } of complex numbers such that f (z ) = n=0 an (z z0 )n , z D(z0 , r).

Suppose that f is analytic on the open upper half plane C+ = {z : Im z > 0}, f is continuous on the closure C+ R of C+ , and Im f (z ) = 0 for z R. Then f has an analytic extension to all of C. Proof. We will give an outline of the argument, leaving the details to the problems at the end of the section. Extend f to a function f dened on C by + f (z ), z C R f (z ) = f (z ), z / C+ R .

2.2.16

Theorem

If f is analytic on , then f is representable in by power series. In fact, if D(z0 , r) , then f (z ) = f (n) (z0 ) (z z0 )n , n ! n=0

z D(z0 , r).

As is the usual practice, we will call this series the Taylor expansion of f about z0 . Proof. Let D(z0 , r) , x any z D(z0 , r), and choose r1 such that |z z0 | < r1 < r. By (2.2.9), Cauchys formula for a circle, f (z ) = Now for w C (z0 , r1 ), (z z0 )n f (w) f (w) 1 f (w) f ( w ) = . = = z z0 wz (w z0 ) (z z0 ) w z0 1 w (w z0 )n+1 z0 n=0 The n-th term of the series has absolute value at most
wC (z0 ,r1 )

1 2i

C (z0 ,r1 )

f (w) dw. wz

max

|f (w)|

|z z0 |n 1 = n+1 r1 r1

wC (z0 ,r1 )

max

|f (w)|

|z z0 | r1

z0 | < 1, the Weierstrass M -test shows that the series converges uniformly on Since |z r1 C (z0 , r1 ). Hence we may integrate term by term, obtaining

f (z ) =
n=0

1 2i

C (z0 ,r1 )

f (n) (z0 ) f (w) n dw ( z z ) = (z z0 )n 0 (w z0 )n+1 n! n=0

Ch: 1

29 (2-15)

TOC

Index

16

CHAPTER 2. THE ELEMENTARY THEORY

by (2.2.11). In order to prove the converse of (2.2.16), namely that a function representable in by power series is analytic on , we need the following basic result.

2.2.17

Theorem

Let {fn } be a sequence of analytic functions on such that fn f uniformly on compact (k ) subsets of . Then f is analytic on , and furthermore, fn f (k) uniformly on compact subsets of for each k = 1, 2, . . . . Proof. First let D(z0 , r) be any closed disk contained in . Then we can choose > r such that D(z0 , ) also. For each z D(z0 , ) and n = 1, 2, . . . , we have, by (2.2.9), fn (z ) = 1 2i f (w) dw. wz

C (z0 ,)

By (2.2.10), f is analytic on D(z0 , ). It follows that f is analytic on . Now by (2.2.11),

(k ) fn (z ) f (k) (z ) =

k! 2i

C (z0 ,)

fn (w) f (w) dw (w z )k+1

and if z is restricted to D(z0 , r), then by the M-L theorem,

(k ) |fn (z ) f (k) (z )|

k! 2

wC (z0 ,)

max

|fn (w) f (w)|

2 0 as n . ( r)k+1
(k )

Thus we have shown that f is analytic on and that fn f (k) uniformly on closed subdisks of . Since any compact subset of can be covered by nitely many closed subdisks, the statement of the theorem follows. The converse of (2.2.16) can now be readily obtained.

2.2.18

Theorem

If f is representable in by power series, then f is analytic on . n Proof. Let D(z0 , r) , and let {an } be such that f (z ) = n=0 an (z z0 ) , z D(z0 , r). By (2.2.7), the series converges uniformly on compact subsets of D(z0 , r), hence by (2.2.17), f is analytic on .

Remark
Since the above series converges uniformly on compact subsets of D(z0 , r), Theorem 2.2.17 also allows us to derive the power series expansion of f (k) from that of f , and to show that the coecients {an } are uniquely determined by z0 and f . For if f (z ) is given by n n=0 an (z z0 ) , z D (z0 , r ), we may dierentiate term by term to obtain

f (k) (z ) =
n=k

n(n 1) (n k + 1)an (z z0 )nk ,

Ch: 1

30 (2-16)

TOC

Index

2.2. POWER SERIES and if we set z = z0 , we nd that ak = f (k) (z0 ) . k!

17

We conclude this section with a result promised in Chapter 1 [see (1.6.1)].

2.2.19

Theorem

If f = u + iv is analytic on , then u and v are harmonic on . v v u Proof. By (1.4.2), f = u x + i x = y i y . But by (2.2.11), f is also analytic on , and thus the Cauchy-Riemann equations for f are also satised. Consequently, x u x = y u y , x v x = y v y .

These partials are all continuous because f is also analytic on .

Problems
1. Prove Theorem 2.2.4. 2. If n z n has radius of convergence r, show that the dierentiated series also has radius of convergence r.
2

nan z n1

3. Let f (x) = e1/x , x = 0; f (0) = 0. Show that f is innitely dierentiable on (, ) and f (n) (0) = 0 for all n. Thus the Taylor series for f is identically 0, hence does not converge to f . Conclude that if r > 0, there is no function g analytic on D(0, r) such that g = f on (r, r). 4. Let {an : n = 0, 1, 2 . . . } be an arbitrary sequence of complex numbers. (a) If lim supn |an+1 /an | = , what conclusions can be drawn about the radius of convergence of the power series n=0 an z n ? (b) If |an+1 /an | approaches a limit , what conclusions can be drawn? 5. If f is analytic at z0 , show that it is not possible that |f (n) (z0 )| > n!bn for all n = 1, 2, . . . , where (bn )1/n as n . 6. Let Rn (z ) be the remainder after the term of degree n in the Taylor expansion of a function f about z0 . (a) Show that Rn (z ) = (z z0 )n+1 2i f (w) dw, (w z )(w z0 )n+1

where = C (z0 , r1 ) as in (2.2.16). (b) If |z z0 | s < r1 , show that |Rn (z )| A(s/r1 )n+1 , where A = Mf ()r1 /(r1 s) and Mf () = max{|f (w)| : w }.

Ch: 1

31 (2-17)

TOC

Index

18

CHAPTER 2. THE ELEMENTARY THEORY

7. (Summation by parts). Let {an } and {bn } be sequences of complex numbers. If bk = bk+1 bk , show that
s s

ak bk = as+1 bs+1 ar br
k=r k=r

bk+1 ak .

8. (a) If {bn } is bounded and the an are real and greater than 0, with a1 a2 0, show that n=1 an bn converges. (b) If bn = bn (z ), that is, the bn are functions from a set S to C, the bn are uniformly bounded on S , and the an are real and decrease to 0 as in (a), show that n=1 an (bn+1 (z ) bn (z )) converges uniformly on S . 9. (a) Show that n=1 z n /n converges when |z | = 1, except at the single point z = 1. (b) Show that n=1 (sin nx)/n converges for real x, uniformly on {x : 2k + x 2(k + 1) }, > 0, k an integer. (c) Show that n=1 (sin nz )/n diverges if x is not real. (The complex sine function will be discussed in the next chapter. It is dened by sin w = (eiw eiw )/2i. 10. Show that the function f occurring in the proof of the Schwarz reection principle is analytic on C \ R and continuous on C. 11. Show that f is analytic on C. 12. Use the following outline to give an alternative proof of the Cauchy integral formula for a circle. (a) Let F (z ) =
C (z0 ,r )

1 dw, z / C (z0 , r). wz

Use (2.2.10), (2.1.6) and (2.1.7b) to show that F is constant on D(z0 , r). (b) F (z0 ) = 2i by direct computation. Theorem 2.2.9 now follows, thus avoiding the series expansion argument that appears in the text. 13. (a) Suppose f is analytic on D(a, r). Prove that for 0 r < R, |f (n) (a)| n! 2rn

|f (a + reit )| dt.

(b) Prove that if f is an entire function such that for some M > 0 and some natural number k , |f (z )| M |z |k for |z | suciently large, then f is a polynomial of degree at most k . (c) Let f be an entire function such that |f (z )| 1 + |z |3/2 for all z . Prove that there are complex numbers a0 , a1 such that f (z ) = a0 + a1 z . 14. Let {an : n = 0, 1, . . . } be a sequence of complex numbers such that n=0 |an | < but n=0 n|an | = . Prove that the radius of convergence of the power series an z n is equal to 1. 15. Let {fn } be a sequence of analytic functions on such that {fn } converges to f uniformly on compact subsets of . Give a proof that f is analytic on , based on

Ch: 1

32 (2-18)

TOC

Index

2.3. THE EXPONENTIAL AND COMPLEX TRIGONOMETRIC FUNCTIONS

19

Moreras theorem [rather than (2.2.10), which was the main ingredient in the proof (k ) of (2.2.17)]. Note that in the present problem we need not prove that fn f (k) uniformly on compact subsets of .

2.3

The Exponential and Complex Trigonometric Functions

In this section, we use our results on power series to complete the discussion of the exponential function and to introduce some of the other elementary functions. Recall (Section 1.5) that exp is dened on C by exp(x + iy ) = ex (cos y + i sin y ); thus exp has magnitude ex and argument y . The function exp satises a long list of properties; for the readers convenience, we give the justication of each item immediately after the statement.

2.3.1

Theorem
n=0

(a) exp is an entire function [this was proved in (1.5.2)]. (b) exp(z ) = z n /n!, z C. Apply (a) and (2.2.16), using the fact [see (1.5.2)] that exp is its own derivative. (c) exp(z1 + z2 ) = exp(z1 ) exp(z2 ). Fix z0 C; for each z C, we have, by (2.2.16), exp(z ) = exp(z0 ) (z z0 )n (z z0 )n = exp(z0 ) = exp(z0 ) exp(z z0 ) by (b). n! n! n=0 n=0

Now set z0 = z1 and z = z1 + z2 . (d) exp has no zeros in C. By (c), exp(z z ) = exp(z ) exp(z ). But exp(z z ) = exp(0) = 1, hence exp(z ) = 0. (e) exp(z ) = 1/ exp(z ) (the argument of (d) proves this also). (f) exp(z ) = 1 i z is an integer multiple of 2i. exp(x + iy ) = 1 i ex cos y = 1 and ex sin y = 0 i ex cos y = 1 and sin y = 0 i x = 0 and y = 2n for some n. (g) | exp(z )| = eRe z (by denition of exp). (h) exp has 2i as a period, and any other period is an integer multiple of 2i. exp(z + w) = exp(z ) i exp(w) = 1 by (c), and the result follows from (f). (i) exp maps an arbitrary vertical line {z : Re z = x0 } onto the circle with center 0 and radius ex0 , and exp maps an arbitrary horizontal line {z : Im z = y0 } one-to-one onto the open ray from 0 through exp(iy0 ). {exp(z ) : Re z = x0 } = {ex0 (cos y + i sin y ) : y R}, which is the circle with center 0 and radius ex0 (covered innitely many times). Similarly, we have {exp(z ) : Im z = y0 } = {ex eiy0 : x R}, which is the desired ray.

Ch: 1

33 (2-19)

TOC

Index

20

CHAPTER 2. THE ELEMENTARY THEORY

(j) For each real number , exp restricted to the horizontal strip {x + iy : y < +2 }, is a one-to-one map onto C \ {0}. This follows from (i) and the observation that as y0 ranges over [, + 2 ), the open rays from 0 through eiy0 sweep out C \ {0}.

Notation
We will often write ez for exp(z ). We now dene sin z and cos z by sin z = eiz eiz , 2i cos z = eiz + eiz . 2

These denitions are consistent with, and are motivated by, the fact that eiy = cos y + i sin y, y R. Since exp is an entire function, it follows from the chain rule that sin and cos are also entire functions and the usual formulas sin = cos and cos = sin hold. Also, it follows from property (f) of exp that sin and cos have no additional zeros in the complex plane, other than those on the real line. (Note that sin z = 0 i eiz = eiz i e2iz = 1.) However, unlike sin z and cos z for real z , sin and cos are not bounded functions. This can be deduced directly from the above denitions, or from Liouvilles theorem, to be proved in the next section. The familiar power series representations of sin and cos hold [and may be derived using (2.2.16)]:

sin z =
n=0

(1)n

z 2n+1 , (2n + 1)!

cos z =
n=0

(1)n

z 2n . (2n)!

Other standard trigonometric functions can be dened in the usual way; for example, tan z = sin z/ cos z . Usual trigonometric identities and dierentiation formulas hold, for d instance, sin(z1 + z2 ) = sin z1 cos z2 + cos z1 sin z2 , dz tan z = sec2 z , and so on. Hyperbolic functions are dened by cosh z = ez + ez , 2 sinh z = ez ez . 2

The following identities can be derived from the denitions: cos iz = cosh z, sin iz = i sinh z

sin(x + iy ) = sin x cosh y + i cos x sinh y,

cos(x + iy ) = cos x cosh y i sin x sinh y.

Also, sinh z = 0 i z = in , n an integer; cosh z = 0 i z = i(2n + 1)/2, n an integer.

Ch: 1

34 (2-20)

TOC

Index

2.4. FURTHER APPLICATIONS

21

Problems
1. Show that for any integer k , sin z maps the strip {x+iy : (2k 1)/2 < x < (2k +1)/2} one-to-one onto C \ {u + iv : v = 0, |u| 1}, and maps {x + iy : x = (2k + 1)/2, y 0} {x + iy : x = (2k 1)/2, y 0} one-to-one onto {u + iv : v = 0, |u| 1}. 2. Find all solutions of the equation sin z = 3. 3. Calculate
C (0,1) sin z z4

dz .

4. Prove that given r > 0, there exists n0 such that if n n0 , then 1+z +z 2 /2!+ +z n /n! has all its zeros in |z | > r. 5. Let f be an entire function such that f + f = 0, f (0) = 0, and f (0) = 1. Prove that f (z ) = sin z for all z C. 6. Let f be an entire function such that f = f and f (0) = 1. What follows and why?

2.4

Further Applications

In this section, we apply the preceding results in a variety of ways. The rst two of these are consequences of the Cauchy integral formula for derivatives (2.2.11).

2.4.1

Cauchys Estimate

Let f be analytic on , and let D(z0 , r) . Then |f (n) (z0 )| n! rn max |f (z )|.

z C (z0 ,r )

Proof. This is immediate from (2.2.11) and the M-L theorem.

Remark
If f (z ) = z n and z0 = 0, we have f (n) (z0 ) = n! = (n!/rn ) maxzC (z0 ,r) |f (z )|, so the above inequality is sharp.

2.4.2

Liouvilles Theorem

If f is a bounded entire function, then f is constant. Proof. Assume that |f (z )| M < for all z C, and x z0 C. By (2.4.1), |f (z0 )| M/r for all r > 0. Let r to conclude that f (z0 ) = 0. Since z0 is arbitrary, f 0, hence f is constant on C by (2.1.7b).

2.4.3

The Fundamental Theorem of Algebra

Suppose P (z ) = a0 + a1 z + + an z n is polynomial of degree n 1. Then there exists z0 C such that P (z0 ) = 0.

Ch: 1

35 (2-21)

TOC

Index

22 Proof. Since |P (z )| = |z |n an +

CHAPTER 2. THE ELEMENTARY THEORY

an1 a0 an + + n |z |n z z 2

for all suciently large |z |, it follows that |P (z )| as z | . If P (z ) is never 0, then 1/P is an entire function. Moreover, |1/P (z )| 0 as |z | , and therefore 1/P is bounded. By (2.4.2), 1/P is constant, contradicting deg P 1. Recall that if P is a polynomial of degree n 1 and P (z0 ) = 0, we may write P (z ) = (z z0 )m Q(z ) where m is a positive integer and Q(z ) is a polynomial (possibly constant) such that Q(z0 ) = 0. In this case P is said to have a zero of order m at z0 . The next denition extends the notion of the order of a zero to analytic functions in general.

2.4.4

Denition

Let f be analytic on and z0 . We say that f has a zero of order m at z0 if there is an analytic function g on such that g (z0 ) = 0 and f (z ) = (z z0 )m g (z ) for all z .

2.4.5

Remark

In terms of the Taylor expansion f (z ) = n=0 an (z z0 )n , f has a zero of order m at z0 i a0 = a1 = = am1 = 0, while am = 0. Equivalently, f (n) (z0 ) = 0 for n = 0, . . . , m 1, while f (m) (z0 ) = 0 (see Problem 2).

2.4.6

Denition

If f : C, the zero set of f is dened as Z (f ) = {z : f (z ) = 0}. Our next major result, the identity theorem for analytic functions, is a consequence of a topological property of Z (f ).

2.4.7

Lemma

Let f be analytic on , and let L be the set of limit points (also called accumulation points or cluster points) of Z (f ) in . Then L is both open and closed in . Proof. First note that L Z (f ) by continuity of f . Also, L is closed in because the set of limit points of any subset of is closed in . (If {zn } is a sequence in L such that zn z , then given r > 0, zn D(z, r) for n suciently large. Since zn is a limit point of Z (f ), D(z, r) contains innitely many points of Z (f ) dierent from zn , and hence innitely many points of Z (f ) dierent from z . Thus z L also.) It remains to show that L is open in . Let z0 L, and write f (z ) = n=0 an (z z0 )n , z D(z0 , r) . Now f (z0 ) = 0, and hence either f has a zero of order m at z0 (for some m), or else an = 0 for all n. In the former case, there is a function g analytic on such that f (z ) = (z z0 )m g (z ), z , with g (z0 ) = 0. By continuity of g , g (z ) = 0 for all z suciently close to z0 , and consequently z0 is an isolated point of Z (f ). But then z0 / L, contradicting out assumption. Thus, it must be the case that an = 0 for all n, so that f 0 on D(z0 , r). Consequently, D(z0 , r) L, proving that L is open in .

Ch: 1

36 (2-22)

TOC

Index

2.4. FURTHER APPLICATIONS

23

2.4.8

The Identity Theorem

Suppose f is analytic on the open connected set . Then either f is identically zero on or else Z (f ) has no limit point in . Equivalently, if Z (f ) has a limit point in , then f is identically 0 on . Proof. By (2.4.7), the set L of limit points of Z (f ) is both open and closed in . Since is connected, either L = , in which case f 0 on , or L = , so that Z (f ) has no limit point in .

2.4.9

Corollary

If f and g are analytic on and {z : f (z ) = g (z )} has a limit point in , then f g . Proof. Apply the identity theorem to f g . Our next application will be to show (roughly) that the absolute value of a function analytic on a set S cannot attain a maximum at an interior point of S . As a preliminary we show that the value of an analytic function at the center of a circle is the average of its values on the circumference.

2.4.10

Theorem
2

Suppose f is analytic on and D(z0 , r) . Then f (z0 ) = 1 2 f (z0 + reit ) dt.

0

Proof. Use (2.2.9), Cauchys integral formula for a circle, with z = z0 . The other preliminary to the proof of the maximum principle is the following fact about integrals.

2.4.11

Lemma

Suppose : [a, b] R is continuous, (t) k for all t, while the average of , namely b 1 ba a (t) dt, is at least k . Then (t) = k for all t. Proof. Observe that
b b

0
a

[k (t)] dt = k (b a)
a

(t) dt 0.

We now consider the maximum principle, which is actually a collection of closely related results rather than a single theorem. We will prove four versions of the principle, arranged in order of decreasing strength.

2.4.12

Maximum Principle

Let f be analytic on the open connected set . (a) If |f | assumes a local maximum at some point in , then f is constant on .

Ch: 1

37 (2-23)

TOC

Index

24

CHAPTER 2. THE ELEMENTARY THEORY

(b) If = sup{|f (z )| : z }, then either |f (z )| < for all z or f is constant on . (c) If is a bounded region and M 0 is such that lim supn |f (zn )| M for each sequence {zn } in that converges to a boundary point of , then |f (z )| < M for all z or f is constant on . (d) Let be a bounded region, with f continuous on the closure of . Denote the boundary of by , and let M0 = max{|f (z )| : z }. Then either |f (z )| < M0 for all z or f is constant on . Consequently, max{|f (z )| : z } = max{|f (z )| : z }. Proof. (a) If |f | assumes a local maximum at z0 , then for some > 0, |f (z )| |f (z0 )| for |z z0 | < . If f (z0 ) = 0, then f (z ) = 0 for all z D(z0 , ), so f 0 by the identity theorem. So assume that f (z0 ) = 0. If 0 < r < , then (2.4.10) with both sides divided by f (z0 ) yields 1= 1 2
2 0

f (z0 + reit ) dt. f (z0 )

Taking the magnitude of both sides, we obtain 1 1 2

2 0

f (z0 + reit ) dt 1 f (z0 )

because |f (z0 + reit )| |f (z0 )| for all t [0, 2 ]. Since this holds for all r (0, ), the preceding lemma (2.4.11) gives |f (z )/f (z0 )| = 1, z D(z0 , ). Now take the real part (rather than the magnitude) of both sides of the above integral, and use the fact that for any complex number w, we have | Re w| |w|. We conclude that Re(f (z )/f (z0 )) = 1 on D(z0 , ). But if |w| = Re w = c, then w = c, hence f (z ) = f (z0 ) on D(z0 , ). By the identity theorem, f is constant on . (b) If = + there is nothing to prove, so assume < +. If |f (z0 )| = for some z0 , then f is constant on by (a). (c) If is dened as in (b), then there is a sequence {zn } in such that |f (zn )| . But since is bounded, there is a subsequence {znj } that converges to a limit z0 . If z0 , then |f (z0 )| = , hence f is constant by (b). On the other hand, if z0 belongs to the boundary of , then M by hypothesis. Again by (b), either |f (z )| < M for all z or f is constant on . (d) Let {zn } be any sequence in converging to a point z0 . Then |f (zn )| |f (z0 )| M0 . By (c), |f | < M0 on or f is constant on . In either case, the maximum of |f | on is equal to the maximum of |f | on . The absolute value of an analytic function may attain its minimum modulus on an open connected set without being constant (consider f (z ) = z on C). However, if the function is never zero, we do have a minimum principle.

2.4.13

Minimum Principle

Let f be analytic and never 0 on the region . (a) If |f | assumes a local minimum at some point in , then f is constant on .

Ch: 1

38 (2-24)

TOC

Index

2.4. FURTHER APPLICATIONS

25

(b) Let = inf {|f (z )| : z }; then either |f (z )| > for all z or f is constant on . (c) If is a bounded region and m 0 is such that lim inf n |f (zn )| m for each sequence {zn } that converges to a boundary point of , then |f (z )| > m for all z or f is constant on . (d) Let be a bounded region, with f is continuous on and m0 = min{|f (z )| : z }. Then either |f (z )| > m0 for all z or f is constant on . As a consequence, we have min{|f (z )| : z } = min{|f (z )| : z }. Proof. Apply the maximum principle to 1/f . Suppose f is analytic on the region , and we put g = ef . Then |g | = eRe f , and hence |g | assumes a local maximum at z0 i Re f has a local maximum at z0 . A similar statement holds for a local minimum. Furthermore, by (2.1.7b), f is constant i f 0 i f ef 0 i g 0 i g is constant on . Thus Re f satises part (a) of both the maximum and minimum principles (note that |g | is never 0). A similar argument can be given for Im f (put g = eif ). Since the real and imaginary parts of an analytic function are, in particular, harmonic functions [see (2.2.19)], the question arises as to whether the maximum and minimum principles are valid for harmonic functions in general. The answer is yes, as we now proceed to show. We will need to establish one preliminary result which is a weak version of the identity theorem (2.4.8) for harmonic functions.

2.4.14

Identity Theorem for Harmonic Functions

If u is harmonic on the region , and u restricted to some subdisk of is constant, then u is constant on . Proof. Let A = {a : u is constant on some disk with center at a}. It follows from the denition of A that A is an open subset of . But \ A is also open; to see this, let z0 \ A and D(z0 , r) . By (1.6.2), u has a harmonic conjugate v on D(z0 , r), so that u is the real part of an analytic function on D(z0 , r). If u is constant on any subdisk of D(z0 , r), then [since u satises (a) of the maximum (or minimum) principle, as indicated in the remarks following (2.4.13)] u is constant on D(z0 , r), contradicting z0 \ A. Thus D(z0 , r) \ A, proving that \ A is also open. Since is connected and A = by hypothesis, we have A = . Finally, x z1 and let B = {z : u(z ) = u(z1 )}. By continuity of u, B is closed in , and since A = , B is also open in . But B is not empty (it contains z1 ), hence B = , proving that u is constant on .

2.4.15

Maximum and Minimum Principle for Harmonic Functions

If u is harmonic on a region and u has either a local maximum or a local minimum at some point of , then u is constant on . Proof. Say u has a local minimum at z0 (the argument for a maximum is similar). Then for some r > 0 we have D(z0 , r) and u(z ) u(z0 ) on D(z0 , r). By (1.6.2) again, u is the real part of an analytic function on D(z0 , r), and we may invoke the minimum principle [as we did in proving (2.4.14)] to conclude that u is constant on D(z0 , r) and hence by (2.4.14), constant on .

Ch: 1

39 (2-25)

TOC

Index

26

CHAPTER 2. THE ELEMENTARY THEORY

Remark
The proof of (2.4.12) shows that part (a) of the maximum principle implies part (b), (b) implies (c), and (c) implies (d), and similarly for the minimum principle. Thus harmonic functions satisfy statements (b), (c) and (d) of the maximum and minimum principles. We conclude this chapter with one of the most important applications of the maximum principle.

2.4.16

Schwarzs Lemma

Let f be analytic on the unit disk D = D(0, 1), and assume that f (0) = 0 and |f (z )| 1 for all z D. Then (a) |f (z )| |z | on D, and (b) |f (0)| 1. Furthermore, if equality holds in (a) for some z = 0, or if equality holds in (b), then f is a rotation of D. That is, there is a constant with || = 1 such that f (z ) = z for all z D. Proof. Dene g (z ) = f (z )/z f (0) if z D \ {0} if z = 0.

By (2.2.13), g is analytic on D. We claim that |g (z )| 1. For if |z | < r < 1, part (d) of the maximum principle yields |g (z )| max{|g (w)| : |w| = r} 1 1 sup{|f (w)| : w D} . r r

Since r may be chosen arbitrarily close to 1, we have |g | 1 on D, proving both (a) and (b). If equality holds in (a) for some z = 0, or if equality holds in (b), then g assumes its maximum modulus at a point of D, and hence g is a constant on D (necessarily || = 1). Thus f (z ) = z for all z D. Schwarzs lemma will be generalized and applied in Chapter 4 (see also Problem 24).

Problems
1. Give an example of a nonconstant analytic function f on a region such that f has a limit point of zeros at a point outside of . 2. Verify the statements made in (2.4.5). 3. Consider the four forms of the maximum principle (2.4.12), for continuous rather than analytic functions. What can be said about the relative strengths of the statements? The proof in the text shows that (a) implies (b) implies (c) implies (d), but for example, does (b) imply (a)? (The region is assumed to be one particular xed open connected set, that is, the statement of the theorem does not have for all in it.) 4. (LHospitals rule). Let f and g be analytic at z0 , and not identically zero in any neighborhood of z0 . If limzz0 f (z ) = limzz0 g (z ) = 0, show that f (z )/g (z ) approaches a limit (possibly ) as z z0 , and limzz0 f (z )/g (z ) = limzz0 f (z )/g (z ). 5. If f is analytic on a region and |f | is constant on , show that f is constant on .

Ch: 1

40 (2-26)

TOC

Index

2.4. FURTHER APPLICATIONS

27

6. Let f be continuous on the closed unit disk D, analytic on D, and real-valued on D. Prove that f is constant. 7. Let f (z ) = sin z . Find max{|f (z )| : z K } where K = {x + iy : 0 x, y 2 }. 8. (A generalization of part (d) of the maximum principle). Suppose K is compact, f is continuous on K , and f is analytic on K , the interior of K . Show that max |f (z )| = max |f (z )|.
z K z K

Moreover, if |f (z0 )| = maxzK |f (z )| for some z0 K , then f is constant on the component of K that contains z0 . 9. Suppose that is a bounded open set (not necessarily connected), f is continuous on and analytic on . Show that max{|f (z )| : z } = max{|f (z )| : z }. 10. Give an example of a nonconstant harmonic function u on C such that u(z ) = 0 for each real z . Thus the disk that appears in the statement of Theorem 2.4.14 cannot be replaced by just any subset of C having a limit point in C. 11. Prove that an open set is connected i for all f, g analytic on , the following holds: If f (z )g (z ) = 0 for every z , then either f or g is identically zero on . (This says that the ring of analytic functions on is an integral domain i is connected.) 12. Suppose that f is analytic on C+ = {z : Im z > 0} and continuous on S = C+ (0, 1). Assume that f (x) = x4 2x2 for all x (0, 1). Show that f (i) = 3. 13. Let f be an entire function such that |f (z )| 1 for all z . Prove that f is constant. 14. Does there exist an entire function f , not identically zero, for which f (z ) = 0 for every z in an uncountable set of complex numbers? 15. Explain why knowing that the trigonometric identity sin( + ) = sin cos + cos sin for all real and implies that the same identity holds for all complex and . 16. Suppose f is an entire function and Im(f (z )) 0 for all z . Prove that f is constant. (Consider exp(if ).) 17. Suppose f and g are analytic and nonzero on D(0, 1), and 2, 3, . . . . Prove that f /g is constant on D(0, 1).
f (1/n) f (1/n)

g (1/n) g (1/n) , n

18. Suppose that f is an entire function, f (0) = 0 and |f (z ) ez sin z | < 4 for all z . Find a formula for f (z ). 19. Let f and g be analytic on D = D(0, 1) and continuous on D. Assume that Re f (z ) = Re g (z ) for all z D. Prove that f g is constant. 20. Let f be analytic on D = D(0, 1). Prove that either f has a zero in D, or there is a sequence {zn } in D such that |zn | 1 and {f (zn )} is bounded. 21. Let u be a nonnegative harmonic function on C. Prove that f is constant. 22. Suppose f is analytic on D(0, 1), f (0) = i, and |f (z )| > 1 whenever |z | = 1. Prove that f has a zero in D(0, 1). 23. Find the maximum value of Re z 3 for z in the unit square [0, 1] [0, 1].

Ch: 1

41 (2-27)

TOC

Index

28

CHAPTER 2. THE ELEMENTARY THEORY

24. Suppose that f is analytic on D(0, 1), with f (0) = 0. Dene fn (z ) = f (z n ) for n = 1, 2, . . . , z D(0, 1). Prove that fn is uniformly convergent on compact subsets of D(0, 1). (Use Schwarzs lemma.) 25. It follows from (2.4.12c) that if f is analytic on D(0, 1) and f (zn ) 0 for each sequence {zn } in D(0, 1) that converges to a point of C (0, 1), then f 0. Prove the following strengthened version for bounded f . Assume only that f (zn ) 0 for each sequence {zn } that converges to a point in some given arc {eit , t } where < , and deduce that f 0. [Hint: Assume without loss of generality that = 0. Then for suciently large n, the arcs Aj = {eit : (j 1) t j }, j = 1, 2, . . . , n cover C (0, 1). Now consider F (z ) = f (z )f (ei z )f (ei2 z ) f (ein z ).] 26. (a) Let be a bounded open set and let {fn } be a sequence of functions that are analytic on and continuous on the closure . Suppose that {fn } is uniformly Cauchy on the boundary of . Prove that {fn } converges uniformly on . If f is the limit function, what are some properties of f ? (b) What complex-valued functions on the unit circle C (0, 1) can be uniformly approximated by polynomials in z ?

Ch: 1

42 (2-28)

TOC

Index

Chapter 3

The General Cauchy Theorem

In this chapter, we consider two basic questions. First, for a given open set , we try to determine which closed paths in have the property that f (z ) dz = 0 for every analytic function f on . Then second, we try to characterize those open sets having the property that f (z ) dz = 0 for all closed paths in and all analytic functions f on . The results, which may be grouped under the name Cauchys theorem, form the cornerstone of analytic function theory. A basic concept in the general Cauchy theory is that of winding number or index of a point with respect to a closed curve not containing the point. In order to make this precise, we need several preliminary results on logarithm and argument functions.

3.1

Logarithms and Arguments

In (2.3.1), property (j), we saw that given a real number , the exponential function when restricted to the strip {x + iy : y < + 2 } is a one-to-one analytic map of this strip onto the nonzero complex numbers. With this in mind, we make the following denition.

3.1.1

Denition

We take log to be the inverse of the exponential function restricted to the strip S = {x + iy : y < + 2 }. We dene arg to the the imaginary part of log . Consequently, log (exp z ) = z for each z S , and exp(log z ) = z for all z C \ {0}. Several important properties of log and arg follow readily from Denition 3.1.1 and the basic properties of exp.

3.1.2

Theorem

(a) If z = 0, then log (z ) = ln |z | + i arg (z ), and arg (z ) is the unique number in [, + 2 ) such that z/|z | = ei arg (z) , in other words, the unique argument of z in [, + 2 ). 1
Ch: 1 2 3 4 5 6 7 43 (3-1) TOC Index

CHAPTER 3. THE GENERAL CAUCHY THEOREM

(b) Let R be the ray [0, ei , ) = {rei : r 0}. The functions log and arg are continuous at each point of the slit complex plane C \ R , and discontinuous at each point of R . (c) The function log is analytic on C \ R , and its derivative is given by log (z ) = 1/z . Proof. (a) If w = log (z ), z = 0, then ew = z , hence |z | = eRe w and z/|z | = ei Im w . Thus Re w = ln |z |, and Im w is an argument of z/|z |. Since Im w is restricted to [, + 2 ) by denition of log , it follows that Im w is the unique argument for z that lies in the interval [, + 2 ). (b) By (a), it suces to consider arg . If z0 C \ R and {zn } is a sequence converging to z0 , then arg (zn ) must converge to arg (z0 ). (Draw a picture.) On the other hand, if z0 R \ {0}, there is a sequence {zn } converging to z0 such that arg (zn ) + 2 = arg (z0 ) = . (c) This follows from Theorem 1.3.2 (with g = exp, 1 = C, f = log , and = C \ R ) and the fact that exp is its own derivative.

3.1.3

Denition

The principal branches of the logarithm and argument functions, to be denoted by Log and Arg, are obtained by taking = . Thus, Log = log and Arg = arg .

Remark
The denition of principal branch is not standardized; an equally common choice for is = 0. Also, having made a choice of principal branch, one can dene wz = exp(z Log w) for z C and w C \ {0}. We will not need this concept, however.

3.1.4

Denition

Let S be a subset of C (or more generally any metric space), and let f : S C \ {0} be continuous. A function g : S C is a continuous logarithm of f if g is continuous on S and f (s) = eg(s) for all s S . A function : S R is a continuous argument of f if is continuous on S and f (s) = |f (s)|ei(s) for all s S .

3.1.5

Examples

(a) If S = [0, 2 ] and f (s) = eis , then f has a continuous argument on S , namely (s) = s + 2k for any xed integer k . (b) If for some , f is a continuous mapping of S into C \ R , then f has a continuous argument, namely (s) = arg (f (s)). (c) If S = {z : |z | = 1} and f (z ) = z , then f does not have a continuous argument on S . Part (a) is a consequence of Denition 3.1.4, and (b) follows from (3.1.4) and (3.1.2b). The intuition underlying (c) is that if we walk entirely around the unit circle, a continuous argument of z must change by 2 . Thus the argument of z must abruptly jump by 2

Ch: 1

44 (3-2)

TOC

Index

3.1. LOGARITHMS AND ARGUMENTS

at the end of the trip, which contradicts continuity. A formal proof will be easier after further properties of continuous arguments are developed (see Problem 3.2.5). Continuous logarithms and continuous arguments are closely related, as follows.

3.1.6

Theorem

Let f : S C be continuous. (a) If g is a continuous logarithm of f , then Im g is a continuous argument of f . (b) If is a continuous argument of f , then ln |f | + i is a continuous logarithm of f . Thus f has a continuous logarithm i f has a continuous argument. (c) Assume that S is connected, and f has continuous logarithms g1 and g2 , and continuous arguments 1 and 2 . Then there are integers k and l such that g1 (s) g2 (s) = 2ik and 1 (s) 2 (s) = 2l for all s S . Thus g1 g2 and 1 2 are constant on S . (d) If S is connected and s, t S , then g (s) g (t) = ln |f (s)| ln |f (t)| + i((s) (t)) for all continuous logarithms g and all continuous arguments of f . Proof. (a) If f (s) = eg(s) , then |f (s)| = eRe g(s) , hence f (s)/|f (s)| = ei Im g(s) as required. (b) If f (s) = |f (s)|ei(s) , then f (s) = eln |f (s)|+i(s) , so ln |f | + i is a continuous logarithm. (c) We have f (s) = eg1 (s) = eg2 (s) , hence eg1 (s)g2 (s) = 1, for all s S . By (2.3.1f), g1 (s) g2 (s) = 2ik (s) for some integer-valued function k . Since g1 and g2 are continuous on S , so is k . But S is connected, so k is a constant function. A similar proof applies to any pair of continuous arguments of f . (d) If is a continuous argument of f , then ln |f | + i is a continuous logarithm of f by part (b). Thus if g is any continuous logarithm of f , then g = ln |f | + i + 2ik by (c). The result follows. As Example 3.1.5(c) indicates, a given zero-free continuous function on a set S need not have a continuous argument. However, a continuous argument must exist when S is an interval, as we now show.

3.1.7

Theorem

Let : [a, b] C \ {0} be continuous, that is, is a curve and 0 / . Then has a continuous argument, hence by (3.1.6), a continuous logarithm. Proof. Let be the distance from 0 to , that is, = min{| (t)| : t [a, b]}. Then > 0 because 0 / and is a closed set. By the uniform continuity of on [a, b], there is a partition a = t0 < t1 < < tn = b of [a, b] such that if 1 j n and t [tj 1 , tj ], then (t) D( (tj ), ). By (3.1.5b), the function , restricted to the interval [t0 , t1 ], has a continuous argument 1 , and restricted to [t1 , t2 ] has a continuous argument 2 . Since 1 (t1 ) and 2 (t1 ) dier by an integer multiple of 2 , we may (if necessary) redene 2 on [t1 , t2 ] so that the relation 1 2 is a continuous argument of on [t0 , t2 ]. Proceeding in this manner, we obtain a continuous argument of on the entire interval [a, b]. For a generalization to other subsets S , see Problem 3.2.6.

Ch: 1

45 (3-3)

TOC

Index

CHAPTER 3. THE GENERAL CAUCHY THEOREM

3.1.8

Denition

Let f be analytic on . We say that g is an analytic logarithm of f if g is analytic on and eg = f . Our next goal is to show that if satises certain conditions, in particular, if is a starlike region, then every zero-free analytic function f on has an analytic logarithm on . First, we give necessary and sucient conditions for f to have an analytic logarithm.

3.1.9

Theorem

Let f be analytic and never zero on the open set . Then f has an analytic logarithm on i the logarithmic derivative f /f has a primitive on . Equivalently, by (2.1.6) (z ) and (2.1.10), f f (z ) dz = 0 for every closed path in . Proof. If g is an analytic logarithm of f , then eg = f , hence f /f = g . Conversely, if f /f has a primitive g , then f /f = g , and therefore (f eg ) = f eg g + f eg = eg (f f g ) which is identically zero on . Thus f eg is constant on f eg = kA on the component A, then kA cannot be zero, so some constant lA . We then have f = eg+lA , so that g + lA is on A. Finally, A (g + lA ) is an analytic logarithm of f on . each component of . If we can write kA = elA for an analytic logarithm of f

We may now give a basic sucient condition on under which every zero-free analytic function on has an analytic logarithm.

3.1.10

Theorem

If is an open set such that h(z ) dz = 0 for every analytic function h on and every closed path in , in particular if is a starlike region, then every zero-free analytic function f on has an analytic logarithm. Proof. The result is a consequence of (3.1.9). If is starlike, then Cauchys theorem for starlike regions (2.1.9).

h(z ) dz = 0 by

3.1.11

Remark

If g is an analytic logarithm of f on , then f has an analytic n-th root, namely f 1/n = exp(g/n). If f (z ) = z and g = log , we obtain z 1/n = exp 1 1 ln |z | + i arg z n n = |z |1/n exp i arg z . n

More generally, we may dene an analytic version of f w for any complex number w, via f w = ewg .

Ch: 1

46 (3-4)

TOC

Index

3.2. THE INDEX OF A POINT WITH RESPECT TO A CLOSED CURVE

3.2

The Index of a Point with Respect to a Closed Curve

In the introduction to this chapter we raised the question of which closed paths in an open set have the property that f (z ) dz = 0 for every analytic function f on . As we will see later, a necessary and sucient condition on is that not wind around any points outside of . That is to say, if z0 / and is dened on [a, b], there is no net change in the argument of (t) z0 as t increases from a to b. To make this precise, we dene the notion of the index (or winding number ) of a point with respect to a closed curve. The following observation will be crucial in showing that the index is well-dened.

3.2.1

Theorem

Let : [a, b] C be a closed curve. Fix z0 / , and let be a continuous argument of z0 [ exists by (3.1.7)]. Then (b) (a) is an integer multiple of 2 . Furthermore, if 1 is another continuous argument of z0 , then 1 (b) 1 (a) = (b) (a). Proof. By (3.1.4), we have ( (t) z0 )/| (t) z0 | = ei(t) , a t b. Since is a closed curve, (a) = (b), hence 1= (b) z0 | (a) z0 | = ei((b)(a)) . | (b) z0 | (a) z0

Consequently, (b) (a) is an integer multiple of 2 . If 1 is another continuous argument of z0 , then by (3.1.6c), 1 = 2l for some integer l. Thus 1 (b) = (b) + 2l and 1 (a) = (a) + 2l, so 1 (b) 1 (a) = (b) (a). It is now possible to dene the index of a point with respect to a closed curve.

3.2.2

Denition

Let : [a, b] C be a closed curve. If z0 / , let z0 be a continuous argument of z0 . The index of z0 with respect to , denoted by n(, z0 ), is n(, z0 ) = z0 (b) z0 (a) . 2

By (3.2.1), n(, z0 ) is well-dened, that is, n(, z0 ) does not depend on the particular continuous argument chosen. Intuitively, n(, z0 ) is the net number of revolutions of (t), a t b, about the point z0 . This is why the term winding number is often used for the index. Note that by the above denition, for any complex number w we have n(, z0 ) = n( + w, z0 + w). If is suciently smooth, an integral representation of the index is available.

3.2.3

Theorem
1 2i 1 dz. z z0

Let be a closed path, and z0 a point not belonging to . Then n(, z0 ) =

Ch: 1

47 (3-5)

TOC

Index

CHAPTER 3. THE GENERAL CAUCHY THEOREM

More generally, if f is analytic on an open set containing , and z0 / (f ) , then n(f , z0 ) = 1 2i f (z ) dz. f (z ) z0

Proof. Let be the distance from z0 to . As in the proof of (3.1.7), there is a partition a = t0 < t1 < < tn = b such that tj 1 t tj implies (t) D( (tj ), ). For each j , z0 / D( (tj ), ) by denition of . Consequently, by (3.1.10), the analytic function z z z0 , when restricted to D( (tj ), ) has an analytic logarithm gj . Now if g is an (z ) = 1/(z z0 ) for all analytic logarithm of f , then g = f /f [see (3.1.9)]. Therefore gj z D( (tj ), ). The path restricted to [tj 1 , tj ] lies in the disk D( (tj ), ), and hence by (2.1.6), 1 dz = gj ( (tj )) gj ( (tj 1 )). z z0

|[tj 1 ,tj ]

Thus 1 dz = z z0
n

[gj ( (tj )) gj ( (tj 1 ))].

j =1

If j = Im gj , then by (3.1.6a), j is a continuous argument of z z z0 on D( (tj ), ). By (3.1.6d), then, 1 dz = z z0

n

[j ( (tj )) j ( (tj 1 ))].

j =1

If is any continuous argument of z0 , then |[tj1 ,tj ] is a continuous argument of ( z0 )|[tj1 ,tj ] . But so is j |[tj1 ,tj ] , hence by (3.1.6c), j ( (tj )) j ( (tj 1 )) = (tj ) (tj 1 ). Therefore, 1 dz = z z0
n

[(tj ) (tj 1 )] = (b) (a) = 2n(, z0 )

j =1

completing the proof of the rst part of the theorem. Applying this result to the path f , we get the second statement. Specically, if z0 / (f ) , then n(f , z0 ) = 1 2i 1 1 dz = z z0 2i f (z ) dz. f (z ) z0

The next result contains additional properties of winding numbers that will be useful later, and which are also interesting (and amusing, in the case of (d)) in their own right.

Ch: 1

48 (3-6)

TOC

Index

3.2. THE INDEX OF A POINT WITH RESPECT TO A CLOSED CURVE

3.2.4

Theorem

Let , 1 , 2 : [a, b] C be closed curves. (a) If z / , then n(, z ) = n( z, 0). (b) If 0 / 1 , then n(1 2 , 0) = n(1 , 0)+ n(2 , 0) and n(1 /2 , 0) = n(1 , 0) n(2 , 0). 2 (c) If D(z0 , r) and z / D(z0 , r), then n(, z ) = 0. (d) If |1 (t) 2 (t)| < |1 (t)|, a t b, then 0 / 1 2 and n(1 , 0) = n(2 , 0). Proof. (a) This follows from Denition 3.2.2. (b) Since 0 / 1 , both n(1 , 0) and 2 , 0) are dened. If 1 and 2 are continuous 2 arguments of 1 and 2 respectively, then j (t) = |j (t)|eij (t) , j = 1, 2, so 1 (t)2 (t) = |1 (t)2 (t)|ei(1 (t)+2 (t)) , Thus n(1 2 , 0) = (1 (b) + 2 (b)) (1 (a) + 2 (a)) = (1 (b) 1 (a)) + (2 (b) 2 (a)) = n(1 , 0) + n(2 , 0). Similarly, n(1 /2 , 0) = n(1 , 0) n(2 , 0). (c) If z / D(z0 , r), then by (3.1.10), the function f dened by f (w) = w z, w D(z0 , r), has an analytic logarithm g . If is the imaginary part of g , then by (3.1.6a), is a continuous argument of z . Consequently, n(, z ) = (2 )1 [( (b)) ( (a))] = 0 since (b) = (a). (d) First note that if 1 (t) = 0 or 2 (t) = 0, then |1 (t) 2 (t)| < |1 (t)| is false; . Let be the closed curve dened by (t) = 2 (t)/1 (t). By the 2 therefore, 0 / 1 hypothesis, we have |1 (t)| < 1 on [a, b], hence D(1, 1). But by (c) and (b), 0 = n(, 0) = n(2 , 0) n(1 , 0). Part (d) of (3.2.4) is sometimes called the dog-walking theorem. (See the text by W. Veech,A Second Course in Complex Analysis, page 30.) For if 1 (t) and 2 (t) are respectively the positions of a man and a dog on a variable length leash, and a tree is located at the origin, then the hypothesis states that the length of the leash is always less than the distance from the man to the tree. The conclusion states that the man and the dog walk around the tree exactly the same number of times. See Problem 4 for a generalization of (d). The nal theorem of this section deals with n(, z0 ) when viewed as a function of z0 . 1 (t)/2 (t) = |1 (t)/2 (t)|ei(1 (t)2 (t)) .

3.2.5

Theorem

If is a closed curve, then the function z n(, z ), z / , is constant on each component of C \ , and is 0 on the unbounded component of C \ . Proof. Let z0 C \ , and choose r > 0 such that D(z0 , r) C \ . If z D(z0 , r), then by parts (a) and (b) of (3.2.4), n(, z ) n(, z0 ) = n( z, 0) n( z0 , 0) = n z ,0 z0 =n 1+ z0 z ,0 . z0

Ch: 1

49 (3-7)

TOC

Index

8 But for each t,

CHAPTER 3. THE GENERAL CAUCHY THEOREM

z0 z r < 1 (t) z0 | (t) z0 | since D(z0 , r) C \ . Therefore the curve 1+(z0 z )/( z0 ) lies in D(1, 1). By part (c) of (3.2.4), n(1+(z0 z )/( z0 ), 0) = 0, so n(, z ) = n(, z0 ). This proves that the function z n(, z ) is continuous on the open set C \ and locally constant. By an argument that we have seen several times, the function is constant on components of C \ . (If z0 C \ and is that component of C \ containing z0 , let A = {z : n(, z ) = n(, z0 )}. Then A is a nonempty subset of and A is both open and closed in , so A = .) To see that n(, z ) = 0 on the unbounded component of C \ , note that D(0, R) for R suciently large. By (3.2.4c), n(, z ) = 0 for z / D(0, R). Since all z outside of D(0, R) belong to the unbounded component of C \ , we are nished.

Problems
1. Suppose is a region in C \ {0} such that every ray from 0 meets . (a) Show that for any R, log is not analytic on . (b) Show, on the other hand, that there exist regions of this type such that z does have an analytic logarithm on . 2. Let f (z ) = (z a)(z b) for z in the region = C \ [a, b], where a and b are distinct complex numbers. Show that f has an analytic square root, but not an analytic logarithm, on . 3. Let f be an analytic zero-free function on . Show that the following are equivalent. (a) f has an analytic logarithm on . (b) f has an analytic k -th root on (that is, an analytic function h such that hk = f ) for every positive integer k . (c) f has an analytic k -th root on for innitely many positive integers k . 4. Prove the following extension of (3.2.4d), the generalized dog-walking theorem. Let 1 , 2 : [a, b] C be closed curves such that |1 (t) 2 (t)| < |1 (t)| + |2 (t)| for all t [a, b]. Prove that n(1 , 0) = n(2 , 0). (Hint: Dene as in the proof of (3.2.4d), and investigate the location of .) Also, what does the hypothesis imply about the dog and the man in this case? 5. Prove the result given in Example 3.1.5(c). 6. Let f be a continuous mapping of the rectangle S = {x + iy : a x b, c y d} into C \ {0}. Show that f has a continuous logarithm. This can be viewed as a generalization of Theorem 3.1.7; to obtain (3.1.7) (essentially), take c = d. 7. Let f be analytic and zero-free on , and suppose that g is a continuous logarithm of f on . Show that g is actually analytic on . 8. Characterize the entire functions f, g such that f 2 + g 2 = 1. (Hint: 1 = f 2 + g 2 = (f + ig )(f ig ), so f + ig is never 0.) 9. Let f and g be continuous mappings of the connected set S into C \ {0}. (a) If f n = g n for some positive integer n, show that f = g exp(i2k/n) for some

Ch: 1

50 (3-8)

TOC

Index

3.3. CAUCHYS THEOREM k = 0, 1, . . . , n 1.Hence if f (s0 ) = g (s0 ) for some s0 S , then f g . (b) Show that C \ {0} cannot be replaced by C in the hypothesis.

3.3

Cauchys Theorem

This section is devoted to a discussion of the global (or homology) version of Cauchys theorem. The elementary proof to be presented below is due to John Dixon, and appeared in Proc. Amer. Math. Soc. 29 (1971), pp. 625-626, but the theorem as stated is originally due to E.Artin.

3.3.1

Cauchys Theorem
f (w) dw = 0;

Let be closed path in such that n(, z ) = 0 for all z C \ . (i) For all analytic functions f on , (ii) If z \ , then n(, z )f (z ) = 1 2i f (w) dw. wz

A path in with n(, z ) = 0 for all z C \ is said to be -homologous to zero. Dixons proof requires two preliminary lemmas.

3.3.2

Lemma

Let f be analytic on , and dene g on by g (w, z ) =

f (w)f (z ) , wz

f (z ),

w=z w = z.

Then g is continuous, and for each xed w , the function given by z g (w, z ) is analytic on . Proof. Let {(wn , zn ), n = 1, 2, . . . } be any sequence in converging to (w, z ) . n )f (zn ) If w = z , then eventually wn = zn , and by continuity of f , g (wn , zn ) = f (w wn zn
f (w)f (z ) wz

= g (w, z ). However, if w = z , then 1 wn zn

[zn ,wn ]

f ( ) d

if wn = zn if wn = zn .

g (wn , zn ) =

f (zn )

In either case, the continuity of f at z implies that g (wn , zn ) f (z ) = g (z, z ). Finally, the function z g (w, z ) is continuous on and analytic on \ {w} (because f is analytic on ). Consequently, z g (w, z ) is analytic on by (2.2.13).

Ch: 1

51 (3-9)

TOC

Index

10

CHAPTER 3. THE GENERAL CAUCHY THEOREM

3.3.3

Lemma

Suppose [a, b] R, and let be a continuous complex-valued function on the product space [a, b]. Assume that for each t [a, b], the function z (z, t) is analytic on . b Dene F on by F (z ) = a (z, t) dt, z . Then F is analytic on and
b

F (z ) =
a

(z, t) dt, z . z

Note that Theorem 2.2.10 on integrals of the Cauchy type is special case of this result. However, (2.2.10) will itself play a part in the proof of (3.3.3). Proof. Fix any disk D(z0 , r) such that D(z0 , r) . Then for each z D(z0 , r),
b

F (z ) =
a

(z, t) dt 1 2i 1 2i
b a C (z0 ,r ) b

= =

(w, t) dw wz (w, t) dt

dt 1 dw wz

(by 2.2.9)

C (z0 ,r )

(Write the path integral as an ordinary denite integral and observe that the interchange in the order of integration is justied by the result that applies to continuous functions on b rectangles.) Now a (w, t) dt is a continuous function of w (to see this use the continuity of on [a, b]), hence by (2.2.10), F is analytic on D(z0 , r) and for each z D(z0 , r), F (z ) = =
a b

1 2i
b

(w, t) dt
C (z0 ,r ) a

1 dw (w z )2

1 2i

C (z0 ,r )

(w, t) dw dt (w z )2

=
a

(z, t) dt z

by (2.2.10) again. Proof of Cauchys Theorem. Let be a closed path in the open set such that n(, z ) = 0 for all z C \ , and let f be an analytic function on . Dene = {z C \ : n(, z ) = 0}. Then C \ , so = C; furthermore, is open by (3.2.5). If z and g is dened as in (3.3.2), then g (w, z ) = (f (w) f (z ))/(w z ) since z / . Thus g (w, z ) dw =

f (w) dw 2in(, z )f (z ) = wz

f (w) dw wz

Ch: 1

52 (3-10)

TOC

Index

3.3. CAUCHYS THEOREM

11

since n(, z ) = 0 for z . The above computation shows that we can dene a function h on C by g (w, z ) dw if z h(z ) = f (w ) dw if z . wz By (2.2.10), h is analytic on , and by (3.3.2) and (3.3.3), h is analytic on . Thus h is an entire function. But for |z | suciently large, n(, z ) = 0 by (3.2.5), hence z . (w ) Consequently, h(z ) = f wz dw 0 as |z | . By Liouvilles theorem (2.4.2), h 0. Now if z \ we have, as at the beginning of the proof, 0 = h(z ) =

g (w, z ) dw =

f (w) dw 2in(, z )f (z ) wz

proving (ii). To obtain (i), choose any z \ and apply (ii) to the function w (w z )f (w), w .

3.3.4

Remarks

Part (i) of (3.3.1) is usually referred to as Cauchys theorem, and part (ii) as Cauchys integral formula. In the above proof we derived (i) from (ii); see Problem 1 for the reverse implication. Also, there is a converse to part (i): If is a closed path in such that f (w) dw = 0 for every f analytic on , then n(, z ) = 0 for every z / . To prove this, take f (w) = 1/(w z ) and apply (3.2.3). It is sometimes convenient to integrate over objects slightly more general than closed paths.

3.3.5

Denitions

Let 1 , 2 , . . . , m be closed paths. If k1 , k2 , . . . , km are integers, then the formal sum = k1 1 + + km m is called a cycle. We dene = m j =1 j , and for any continuous function f on ,
m

f (w) dw =
j =1

kj
j

f (w) dw.

Finally, for z / , dene

m

n(, z ) =
j =1

kj n(j , z ).

It follows directly from the above denitions that the integral representation (3.2.3) for winding numbers extends to cover cycles as well. Also, the proof of Cauchys theorem (3.3.1) may be repeated almost verbatim for cycles (Problem 2). Cauchys theorem, along with the remarks and denitions following it combine to yield the following equivalence.

Ch: 1

53 (3-11)

TOC

Index

12

CHAPTER 3. THE GENERAL CAUCHY THEOREM

3.3.6

Theorem

Let be a closed path (or cycle) in the open set . Then function f on i n(, z ) = 0 for every z / .

f (z ) dz = 0 for every analytic

Proof. Apply (3.3.1), (3.3.4) and (3.3.5). Note that the proof of the converse of (i) of (3.3.1) given in (3.3.4) works for cycles, because the integral representation (3.2.3) still holds.

3.3.7

Corollary

Let 1 and 2 be closed paths (or cycles) in the open set . Then 1 f (w) dw = f (w) dw for every analytic function f on i n(1 , z ) = n(2 , z ) for every z / . 2 Proof. Apply (3.3.6) to the cycle 1 2 . Note that Theorem 3.3.6 now provides a solution of the rst problem posed at the beginning of the chapter, namely, a characterization of those closed paths in such that f (z ) dz = 0 for every analytic function f on .

Problems
1. Show that (i) implies (ii) in (3.3.1). 2. Explain briey how the proof of (3.3.1) is carried out for cycles. 3. Let , and f be as in (3.3.1). Show that for each k = 0, 1, 2, . . . and z \ , we have n(, z )f (k) (z ) = 4. Compute
1 C (0,2) z 2 1

k! 2i

f (w) dw. (w z )k+1

dz .
j ez +cos z z4

5. Use Problem 3 to calculate each of the integrals are the paths indicated in Figure 3.3.1.

dz, j = 1, 2, where the j

6. Consider : [0.2 ] C given by (t) = a cos t + ib sin t, where a and b are nonzero real numbers. Evaluate dz/z , and using this result, deduce that
2 0

2 dt = . ab a2 cos2 t + b2 sin2 t

3.4

Another Version of Cauchys Theorem

In this section we consider the second question formulated at the beginning of the chapter: Which open sets have the property that f (z ) dz = 0 for all analytic functions f on and all closed paths (or cycles) in ? A concise answer is given by Theorem 3.4.6, but several preliminaries are needed.

Ch: 1

54 (3-12)

TOC

Index

3.4. ANOTHER VERSION OF CAUCHYS THEOREM

13

Figure 3.3.1

3.4.1

The Extended Complex Plane

2 3 2 Let S = {(x1 , x2 , x3 ) R3 : x2 1 + x2 +(x3 1/2) = 1/4}. Thus S is the sphere in R (called the Riemann sphere ) with center at (0,0,1/2) and radius 1/2 (Figure 3.4.1). The line segment joining (0,0,1), the north pole of S , to a point (x, y, 0) is {(tx, ty, 1t) : 0 t 1}, and this segment meets S when and only when

1 1 t2 (x2 + y 2 ) + ( t)2 = , 2 4

or t =

1 . 1 + x2 + y 2

Therefore the intersection point is (x1 , x2 , x3 ), where x1 = x y x2 + y 2 , x2 = , x3 = . 2 2 2 2 1+x +y 1+x +y 1 + x2 + y 2 (1)

Since 1 x3 = 1/(1 + x2 + y 2 ), it follows from (1) that x2 x1 , y= . x= 1 x3 1 x3

(2)

Let h be the mapping that takes (x, y, 0) to the point (x1 , x2 , x3 ) of S . Then h maps R2 {0}, which can be identied with C, one-to-one onto S \ {(0, 0, 1)}. Also, by (2), x1 x2 h1 (x1 , x2 , x3 ) = ( 1 x3 , 1x3 , 0). Consequently, h is a homeomorphism, that is, h and h1 are continuous. We can identify C and S \ {(0, 0, 1)} formally as follows. Dene k : C R2 {0}. by k (x + iy ) = (x, y.0). Then k is an isometry (a one-to-one, onto, distance-preserving map), hence h k is a homeomorphism of C onto S \ {(0, 0, 1)}. Next let denote a point not to be C {}. Dene g : C S by belonging to C, and take C g (z ) = h(k (z )), z C (0, 0, 1), z = .

is Then g maps C one-to-one onto S . If is the usual Euclidean metric of R3 and d dened on C C by (z, w) = (g (z ), g (w)), d

Ch: 1

55 (3-13)

TOC

Index

14

CHAPTER 3. THE GENERAL CAUCHY THEOREM

N (0,0,1)

(0,0,1/2)

(x 1,x2 ,x3 ) 0 (x,y,0)

Figure 3.4.1 is a metric on C -distance between points of C . (The d is the Euclidean distance then d ) is , d between the corresponding points on the Riemann sphere.) The metric space (C is called the chordal metric on C . It is a consequence of called the extended plane, and d that (C ) and (S, ) are isometric spaces. The following formulas for , d the denition of d hold. d

3.4.2

Lemma
|z w| (1+|z|2 )1/2 (1+|w|2 )1/2 ,
1 , (1+|z |2 )1/2

z, w C z C , w = .

(z, w) = d

Proof. Suppose z = x + iy, w = u + iv . Then by (1) of (3.4.1), (z, w)]2 = [d x y |z |2 u v |w|2 + + 1 + |z |2 1 + |w|2 1 + |z |2 1 + |w|2 1 + |z |2 1 + |w|2 2 2 4 2 2 2 2 4 u + v + |w| xu + yv + |z | |w| x + y + |z | + 2 = 2 2 2 2 (1 + |z | ) (1 + |w| ) (1 + |z |2 )(1 + |w|2 ) |z |2 |w|2 |z |2 + |w|2 |z w|2 + 2|z |2 |w|2 = + 2 2 1 + |z | 1 + |w| (1 + |z |2 )(1 + |w|2 ) |z w|2 = (1 + |z |2 )(1 + |w|2 )
2 2 2

Ch: 1

56 (3-14)

TOC

Index

3.4. ANOTHER VERSION OF CAUCHYS THEOREM as desired. Also, (z, )]2 = [(g (z ), (0, 0, 1))]2 [d = x2 + y 2 + 1 by (1) of (3.4.1) (1 + x2 + y 2 )2 1 . = 1 + |z |2

15

. Here is a list of the most basic properties of C

3.4.3

Theorem

) is compact, and the identity function on C is a homeomorphism , d (a) The metric space (C ). of C (with the usual metric) onto (C, d . In fact, a sequence {zn } in C converges (b) The complex plane is a dense subspace of C to i {|zn |} converges to +. ) is connected and complete. , d (c) The metric space (C (d) Let be a closed curve in C, and dene n(, ) = 0. Then the function n(, ) is \ . continuous on C ) with the one-point compacti is a homeomorphism of (C , d (e) The identity map on C cation (C , T ) of C. (Readers unfamiliar with the one-point compactication of a locally compact space may simply ignore this part of the theorem, as it will not be used later.) ). The , d Proof. Since the Riemann sphere is compact, connected and complete, so is (C in (3.4.2) shows that the identity map on C is a homeomorphism of C into formula for d , and that zn i |zn | +. This proves (a), (b) and (c). Part (d) follows from C (3.2.5). For (e), see Problem 4. We are now going to make precise, in two equivalent ways, the notion that an open set has no holes.

3.4.4

Theorem

\ is connected i each closed curve (and each cycle) in Let be open in C. Then C is -homologous to 0, that is, n(, z ) = 0 for all z / . \ is connected, and let be a closed curve in . Since Proof. Suppose rst that C z n(, z ) is a continuous integer-valued function on C \ [by (3.2.5) and (3.4.3d)], it \ . But n(, ) = 0, hence n(, z ) = 0 for all must be constant on the connected set C \ . The statement for cycles now follows from the result for closed curves. zC The converse is considerably more dicult, and is a consequence of what we will call the hexagon lemma. As we will see, this lemma has several applications in addition to its use in the proof of the converse.

Ch: 1

57 (3-15)

TOC

Index

16

CHAPTER 3. THE GENERAL CAUCHY THEOREM

3.4.5

The Hexagon Lemma

Let be an open subset of C, and let K be a nonempty compact subset of . Then there are closed polygonal paths 1 , 2 , . . . , m in \ K such that
m

n(j , z ) =
j =1

1 if z K 0 if z / .

The lemma may be expressed by saying there is a (polygonal) cycle in \ K which winds around each point of K exactly once, but does not wind around any point of C \ . Proof. For each positive integer n, let Pn be the hexagonal partition of C determined by the hexagon with base [0, 1/n]; see Figure 3.4.2. Since K is a compact subset of the open

0 H

1/n

Figure 3.4.2. A Hexagonal Partition of C. set , we have dist(K, C \ ) > 0, and therefore we can choose n large enough so that if H Pn and H K = , then H . Dene K = {H Pn : H K = }. Since K is nonempty and bounded, K is a nonempty nite collection and K {H : H K} . Now assign a positive (that is, counterclockwise) orientation to the sides of each hexagon (see Figure 3.4.2). Let S denote the collection of all oriented sides of hexagons in K that are sides of exactly one member of K. Observe that given an oriented side ab S , there are unique oriented sides ca and bd in S . (This uniqueness property is the motivation for tiling with hexagons instead of squares. If we used squares instead, as in Figure 3.4.3, we have ab, bc and bd S , so ab does not have a unique successor, thus complicating the argument that follows.) By the above observations, and the fact that S is a nite collection, it follows that given a1 a2 S , there is a uniquely dened closed polygonal path 1 = [a1 , a2 , . . . , ak , a1 ] with all sides in S . If S1 consists of the edges of 1 and S \ S1 = , repeat the above

Ch: 1

58 (3-16)

TOC

Index

3.4. ANOTHER VERSION OF CAUCHYS THEOREM

17

K = shaded areas

Figure 3.4.3 Nonuniqueness when squares are used.

construction with S replaced by S \ S1 . Continuing in this manner, we obtain pairwise disjoint collections S1 , S2 , . . . , Sn such that S = m j =1 Sj , and corresponding closed polygonal paths 1 , 2 , . . . , m (Figure 3.4.4). Suppose now that the hexagons in K are H1 , H2 , . . . , Hp , and let j denote the boundary of Hj , oriented positively. If z belongs to the interior of some Hr , then n(r , z ) = 1 and n(j , z ) = 0, j = r. Consequently, n(1 + 2 + + p , z ) = 1 by (3.3.5). But by construction, n(1 + + m , z ) = n(1 + + p , z ). (The key point is that if both hexagons containing a particular side [a, b] belong to K, then ab / S and ba / S . Thus [a, b] will not contribute to either n(1 + + m , z ) or to n(1 + + p , z ). If only one hexagon containing [a, b] belongs to K, then ab (or ba) appears in both cycles.) Therefore n(1 + + m , z ) = 1. Similarly, if z / , then n(1 + + m , z ) = n(1 + + p , z ) = 0. Finally, assume z K and z belongs to a side s of some Hr . Then s cannot be in S , so z / (1 + + m ) . Let {wk } be a sequence of interior points of Hr with wk converging to z . We have shown that n(1 + + m , wk ) = 1 for all k , so by (3.2.5), n(1 + + m , z ) = 1.

Completion of the Proof of (3.4.4)

\ is not connected, we must exhibit a cycle in that is not -homologous to 0. Now If C is closed and not connected, it can be expressed as the union of two nonempty since C disjoint closed sets K and L. One of these two sets must contain ; assume that L. Then K must be a compact subset of the complex plane C, and K is contained in the plane open set 1 = C \ L. apply the hexagon lemma (3.4.5) to 1 and K to obtain a cycle in 1 \ K = C \ (K L) = such that n(, z ) = 1 for each z K (and n(, z ) = 0 for z / 1 ). Pick any point z in the nonempty set K C \ . Then z / and n(, z ) = 1 = 0.

Ch: 1

59 (3-17)

TOC

Index

18

CHAPTER 3. THE GENERAL CAUCHY THEOREM

Figure 3.4.4. Construction of the closed paths.

Remark
\ is not connected, As a consequence of the denition (3.3.5) of the index of a cycle, if C there must actually be a closed path in such that n(, z ) = 0 for some z / . The list of equivalences below is essentially a compilation of results that have already been established.

3.4.6

Second Cauchy Theorem

Let be an open subset of C. The following are equivalent. \ is connected. (1) C (2) n(, z ) = 0 for each closed path (or cycle) in and each point z C \ . (3) f (z ) dz = 0 for every closed path (or cycle) in and every analytic function f on . (4) Every analytic function on has a primitive on . (5) Every zero-free analytic function on has an analytic logarithm. (6) Every zero-free analytic function on has an analytic n-th root for n = 1, 2, . . . . Proof. (1) is equivalent to (2) by Theorem 3.4.4. (2) is equivalent to (3) by Theorem 3.3.6. (3) is equivalent to (4) by Theorems 2.1.6 and 2.1.10. (3) implies (5) by Theorem 3.1.10. (5) is equivalent to (6) by Problem 3.2.3. (5) implies (2): If z0 / , let f (z ) = z z0 , z . Then f has an analytic logarithm on , and hence for each closed path (or cycle) in we have, by (3.2.3) and (3.1.9), n(, z0 ) = 1 2i 1 1 dz = z z0 2i f (z ) dz = 0. f (z )

Ch: 1

60 (3-18)

TOC

Index

3.4. ANOTHER VERSION OF CAUCHYS THEOREM

19

We will be adding to the above list in later chapters. An open subset of C satisfying any (and hence all) of the conditions of (3.4.6) is said to be (homologically) simply connected. It is true that in complex analysis, the implications (1) (2) (3) are used almost exclusively. The rather tedious hexagon lemma was required to establish the reverse implication (2) (1). Thus one might wonder why we have gone to the trouble of obtaining the hexagon lemma at all. One answer is that it has other applications, including the following global integral representation formula. This formula should be compared with Cauchys integral formula for a circle (2.2.9). It will also be used later in the proof of Runges theorem on rational approximation.

3.4.7

Theorem

Let K be a compact subset of the open set . Then there is a cycle in \ K such that is a formal sum of closed polygonal paths, and for every analytic function f on , f (z ) = 1 2i f (w) dw = 0 for all z K. wz

Proof. Apply the hexagon lemma and part (ii) of (3.3.1).

Problems
1. (a) Give an example of an open connected set that is not simply connected. For this set, describe explicitly an analytic function f and a closed path such that f (z ) dz = 0. (b) Give an example of an open, simply connected set that is not connected. 2. Suppose that in the hexagon lemma, is assumed to be connected. Can a cycle that satises the conclusion be taken to be a closed path ? 3. Let 1 be the ray [1, i/2, ) = {1 t + ti/2 : 0 t < } and let 2 be the ray [1, 2, ). (a) Show that 1 z has analytic square roots f and g on C \ 1 and C \ 2 respectively, such that f (0) = g (0) = 1. (b) Show that f = g below = 1 2 and f = g above . (Compare Problem 3.2.9.) (c) Let h(z ) be given by the binomial expansion of (1 z )1/2 , that is,

h(z ) =
n=0

1/2 (z )n , n

|z | < 1,

where

w n

w(w1)(wn+1) . n!

What is the relationship between h and f ?

4. Prove Theorem 3.4.3(e).

Ch: 1

61 (3-19)

TOC

Index

Chapter 4

Applications Of The Cauchy Theory

This chapter contains several applications of the material developed in Chapter 3. In the rst section, we will describe the possible behavior of an analytic function near a singularity of that function.

4.1

Singularities

We will say that f has an isolated singularity at z0 if f is analytic on D(z0 , r) \ {z0 } for some r. What, if anything, can be said about the behavior of f near z0 ? The basic tool needed to answer this question is the Laurent series, an expansion of f (z ) in powers of z z0 in which negative as well as positive powers of z z0 may appear. In fact, the number of negative powers in this expansion is the key to determining how f behaves near z0 . From now on, the punctured disk D(z0 , r) \ {z0 } will be denoted by D (z0 , r). We will need a consequence of Cauchys integral formula.

4.1.1

Theorem

Let f be analytic on an open set containing the annulus {z : r1 |z z0 | r2 }, 0 < r1 < r2 < , and let 1 and 2 denote the positively oriented inner and outer boundaries of the annulus. Then for r1 < |z z0 | < r2 , we have f (z ) = 1 2i 1 f (w) dw wz 2i f (w) dw. wz

Proof. Apply Cauchys integral formula [part (ii) of (3.3.1)] to the cycle 2 1 . 1
Ch: 1 2 3 4 5 6 7 62 (4-1) TOC Index

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

4.1.2

Denition

For 0 s1 < s2 + and z0 C, we will denote the open annulus {z : s1 < |z z0 | < s2 } by A(z0 , s1 , s2 ).

4.1.3

Laurent Series Representation

If f is analytic on = A(z0 , s1 , s2 ), then there is a unique two-tailed sequence {an } n= such that

f (z ) =
n=

an (z z0 )n , z .

In fact, if r is such that s1 < r < s2 , then the coecients an are given by an = 1 2i f (w) dw, n = 0, 1, 2, . . . . (w z0 )n+1

C (z0 ,r )

Also, the above series converges absolutely on and uniformly on compact subsets of . Proof. Choose r1 and r2 such that s1 < r1 < r2 < s2 and consider the Cauchy type integral 1 2i f (w) dw, z D(z0 , r2 ). wz

C (z0 ,r2 )

Then proceeding just as we did in the proof of Theorem 2.2.16, we obtain 1 2i where an = 1 2i
C (z0 ,r2 )

C (z0 ,r2 )

f (w) an (z z0 )n dw = wz n=0 f (w) dw. (w z0 )n+1

The series converges absolutely on D(z0 , r2 ), and uniformly on compact subsets of D(z0 , r). Next, consider the Cauchy type integral This can be written as 1 2i f (w) (z z0 )[1
wz0 z z0 ]

1 2i

C (z0 ,r1 )

f (w) dw, |z z0 | > r1 . wz 1 2i

dw =

f (w)
C (z0 ,r1 ) n=1

C (z0 ,r1 )

(w z0 )n1 (z z0 )n

dw.

By the Weierstrass M -test, the series converges absolutely and uniformly for w C (z0 , r1 ). Consequently, we may integrate term by term to obtain the series
n=1

bn (z z0 )n , where bn =

1 2i

C (z0 ,r1 )

f (w) dw. (w z0 )n+1

Ch: 1

63 (4-2)

TOC

Index

4.1. SINGULARITIES

This is a power series in 1/(z z0 ), and it converges for |z z0 | > r1 , and hence uniformly on sets of the form {z : |z z0 | 1/} where (1/) > r1 . It follows that the convergence is uniform on compact (indeed on closed) subsets of {z : z z0 | > r1 . The existence part of the theorem now follows from (4.1.1) and the above computations, if we note two facts. First, if s1 < r < s2 and k = 0, 1, 2, . . . ,
C (z0 ,r )

f (w) dw = (w z0 )k+1

C (z0 ,r1 )

f (w) dw = (w z0 )k+1

C (z0 ,r2 )

f (w) dw. (w z0 )k+1

Second, any compact subset of A(z0 , s1 , s2 ) is contained in {z : 1 |z z0 | 2 } for some 1 and 2 with s1 < 1 < 2 < s2 . We turn now to the question of uniqueness. Let {bn } be a sequence such that f (z ) = n n= bn (z z0 ) for z A(z0 , s1 , s2 ). As in the above argument, this series must converge uniformly on compact subsets of A(z0 , s1 , s2 ). Therefore if k is any integer and s1 < r < s2 , then 1 2i f (w) 1 dw = k +1 (w z0 ) 2i =
n= C (z0 ,r ) n=

C (z0 ,r )

bn (w z0 )nk1 dw (w z0 )nk1 dw

bn

1 2i

C (z0 ,r )

= bk , because 1 2i (w z0 )nk1 dw = 1 if n k 1 = 1 0 otherwise.

C (z0 ,r )

The theorem is completely proved. We are now in a position to analyze the behavior of f near an isolated singularity. As the preceding discussion shows, if f has an isolated singularity at z0 , then f can be represented uniquely by

f (z ) =
n=

an (z z0 )n

in some deleted neighborhood of z0 .

4.1.4

Denition

Suppose f has an isolated singularity at z0 , and let n= an (z z0 )n be the Laurent expansion of f about z0 , that is, the series given in (4.1.3). We say that f has a removable singularity at z0 if an = 0 for all n < 0; f has a pole of order m at z0 if m is the largest positive integer such that am = 0. (A pole of order 1 is called a simple pole.) Finally, if an = 0 for innitely many n < 0, we say that f has an essential singularity at z0 . The next theorem relates the behavior of f (z ) for z near z0 to the type of singularity that f has at z0 .

Ch: 1

64 (4-3)

TOC

Index

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

4.1.5

Theorem

Suppose that f has an isolated singularity at z0 . Then (a) f has a removable singularity at z0 i f (z ) approaches a nite limit as z z0 i f (z ) is bounded on the punctured disk D (z0 , ) for some > 0. (b) For a given positive integer m, f has a pole of order m at z0 i (z z0 )m f (z ) approaches a nite nonzero limit as z z0 . Also, f has a pole at z0 i |f (z )| + as z z0 . (c) f has an essential singularity at z0 i f (z ) does not approach a nite or innite limit as z z0 . as z z0 , that is, f (z ) has no limit in C n Proof. Let {an } n= and r > 0 be such that f (z ) = an (z z0 ) for 0 < |z z0 | < r . (a) If an = 0 for all n < 0, then limzz0 f (z ) = a0 . Conversely, if limzz0 f (z ) exists (in C), then f can be dened (or redened) at z0 so that f is analytic on D(z0 , r). It follows n that there is a sequence {bn } n=0 such that f (z ) = n=0 bn (z z0 ) for z D (z0 , r ). By uniqueness of the Laurent expansion, we conclude that an = 0 for n < 0 and an = bn for n 0. (Thus in this case, the Laurent and Taylor expansions coincide.) The remaining equivalence stated in (a) is left as an exercise (Problem 1). (b) If f has a pole of order m at z0 , then for 0 < |z z0 | < r,

f (z ) = am (z z0 )m + + a1 (z z0 )1 +

n=0

an (z z0 )n

where am = 0. Consequently, (z z0 )m f (z ) am = 0 as z z0 . Conversely, if limzz0 (z z0 )m f (z ) = 0, then by (a) applied to (z z0 )m f (z ), there is a sequence {bn } n=0 such that

(z z0 ) f (z ) =

n=0

bn (z z0 )n , z D (z0 , r).

Let z z0 to obtain b0 = limzz0 (z z0 )m f (z ) = 0. Thus f (z ) can be written as b0 (z z0 )m + b1 (z z0 )m+1 + , showing that f has a pole of order m at z0 . The remaining equivalence in (b) is also left as an exercise (Problem 1). as z z0 , then by (a) and (b), f must have an (c) If f (z ) does not have a limit in C essential singularity at z0 . Conversely, if f has an essential singularity at z0 , then (a) and . (b) again imply that limzz0 f (z ) cannot exist in C The behavior of a function near an essential singularity is much more pathological even than (4.1.5c) suggests, as the next theorem shows.

4.1.6

Casorati-Weierstrass Theorem

Let f have an isolated essential singularity at z0 . Then for any complex number w, f (z ) comes arbitrarily close to w in every deleted neighborhood of z0 . That is, for any > 0, f (D (z0 , )) is a dense subset of C. Proof. Suppose that for some > 0, f (D (z0 , )) is not dense in C. Then for some w C, there exists > 0 such that D(w, ) does not meet f (D (z0 , )). For z D (z0 , ), put

Ch: 1

65 (4-4)

TOC

Index

4.1. SINGULARITIES

g (z ) = 1/(f (z ) w)). Then g is bounded and analytic on D (z0 , ), and hence by (4.1.5a), g has a removable singularity at z0 . Let m be the order of the zero of g at z0 (set m = 0 if g (z0 ) = 0) and write g (z ) = (z z0 )m g1 (z ) where g1 is analytic on D(z0 , ) and g1 (z0 ) = 0 [see (2.4.4)]. Then (z z0 )m g1 (z ) = 1/(f (z ) w), so as z approaches z0 , (z z0 )m f (z ) = (z z0 )m w + 1 g1 (z ) w + 1/g1 (z0 ) if m = 0 1/g1 (z0 ) if m = 0.

Thus f has a removable singularity or a pole at z0 .

4.1.7

Remark

The Casorati-Weierstrass theorem is actually a weak version of a much deeper result called the big Picard theorem, which asserts that if f has an isolated essential singularity at z0 , then for any > 0, f (D (z0 , )) is either the complex plane C or C minus one point. We will not prove this result. The behavior of a complex function f at may be studied by considering g (z ) = f (1/z ) for z near 0. This allows us to talk about isolated singularities at . Here are the formal statements.

4.1.8

Denition

We say that f has an isolated singularity at if f is analytic on {z : |z | > r} for some r; thus the function g (z ) = f (1/z ) has an isolated singularity at 0. The type of singularity of f at is then dened as that of g at 0.

4.1.9

Remark

Liouvilles theorem implies that if an entire function f has a removable singularity at , then f is constant. (By (4.1.5a), f is bounded on C.)

Problems
1. Complete the proofs of (a) and (b) of (4.1.5). (Hint for (a): If f is bounded on D (z0 , ), consider g (z ) = (z z0 )f (z ).) 2. Classify the singularities of each of the following functions (include the point at ). (a) z/ sin z (b) exp(1/z ) (c) z cos 1/z (d) 1/[z (ez 1)] (e) cot z

3. Obtain three dierent Laurent expansions of (7z 2)/z (z + 1)(z 2) about z = 1. (Use partial fractions.)

4. Obtain all Laurent expansions of f (z ) = z 1 + (z 1)2 + (z + 2)1 about z = 0, and indicate where each is valid. 5. Find the rst few terms in the Laurent expansion of
1 z 2 (ez ez )

6. Without carrying out the computation in detail, indicate a relatively easy procedure for nding the Laurent expansion of 1/ sin z valid for < |z | < 2 .

valid for 0 < |z | < .

Ch: 1

66 (4-5)

TOC

Index

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY 7. (Partial Fraction Expansion). Let R(z ) = P (z )/Q(z ), where P and Q are polynomials and deg P < deg Q. (If this is not the case, then by long division we may write P (z )/Q(z ) = an z n + + a1 z + a0 + P1 (z )/Q(z ) where deg P1 < deg Q.) Suppose that the zeros of Q are at z1 , . . . , zk with respective orders n1 , . . . , nk . Show that k R(z ) = j =1 Bj (z ), where Bj (z ) is of the form Aj,(nj 1) Aj,0 + + , n (z zj ) j (z zj ) with Aj,r = lim
r

z zj

1 dr [(z zj )nj R(z )] r! dz r

d ( dz r f (z ) is interpreted as f (z ) when r = 0).

Apply this result to R(z ) = 1/[z (z + i)3 ].

n 8. Find the sum of the series sin nz (in closed form), and indicate where n=0 e the series converges. Make an appropriate statement about uniform convergence. (Suggestion: Consider n=0 en einz and n=0 en einz .) , then f is constant. 9. (a) Show that if f is analytic on C (b) Suppose f is entire and there exists M > 0 and k > 0 such that |f (z )| M |z |k for |z | suciently large. Show that f (z ) is a polynomial of degree at most k . (This can also be done without series; see Problem 2.2.13.) (c) Prove that if f is entire and has a nonessential singularity at , then f is a polynomial. (that is, any singularity of f in C is a pole), (d) Prove that if f is meromorphic on C then f is a rational function.

10. Classify the singularities of the following functions (include the point at ). (a) sin2 z z4 (b) 1 1 + sin z 2 (z + 1) z (c) csc z k z 1 (d) exp(tan ) z (e) 1 . sin(sin z )

11. Suppose that a and b are distinct complex numbers. Show that (z a)/(z b) has an analytic logarithm on C \ [a, b], call it g . Then nd the possible Laurent expansions of g (z ) about z = 0. 12. Suppose f is entire and f (C) is not dense in C. Show that f is constant. 13. Assume f has a pole of order m at , and P is a polynomial of degree n. Prove that the composition P f has a pole of order mn at .

4.2

Residue Theory

We now develop a technique that often allows for the rapid evaluation of integrals of the form f (z ) dz , where is a closed path (or cycle) in and f is analytic on except possibly for isolated singularities.

Ch: 1

67 (4-6)

TOC

Index

4.2. RESIDUE THEORY

4.2.1

Denition

Let f have an isolated singularity at z0 , and let the Laurent expansion of f about z0 be n n= an (z z0 ) . The residue of f at z0 , denoted by Res(f, z0 ), is dened to be a1 .

4.2.2

Remarks

In many cases, the evaluation of an integral can be accomplished by the computation of residues. This is illustrated by (a) and (b) below. (a) Suppose f has an isolated singularity at z0 , so that f is analytic on D (z0 , ) for some > 0. Then for any r such that 0 < r < , we have f (w) dw = 2i Res(f, z0 ).
C (z0 ,r )

Proof. Apply the integral formula (4.1.3) for a1 . (b) More generally, if is a closed path or cycle in D (z0 , ) such that n(, z0 ) = 1 and n(, z ) = 0 for every z / D(z0 , ), then f (w) dw = 2i Res(f, z0 ).

Proof. This follows from (3.3.7). (c) Res(f, z0 ) is that number k such that f (z ) [k/(z z0 )] has a primitive on D (z0 , ). Proof. Note that if 0 < r < , then by (a), f (w) k w z0 dw = 2i[Res(f, z0 ) k ].

C (z0 ,r )

Thus if f (z ) [k/(z z0 )] has a primitive on D (z0 , ), then the integral is zero, and hence Res(f, z0 ) = k . Conversely, if Res(f, z0 ) = k , then f (z ) k = an (z z0 )n , z z0 n=
n=1

which has a primitive on D (z0 , ), namely an (z z0 )n+1 . n +1 n=

n=1

(d) If f has a pole of order m at z0 , then Res(f, z0 ) = 1 lim (m 1)! zz0 dm1 [(z z0 )m f (z )] . dz m1

Ch: 1

68 (4-7)

TOC

Index

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

In particular, if f has a simple pole at z0 , then Res(f, z0 ) = lim [(z z0 )f (z )].

z z0

Proof. Let {an } be the Laurent coecient sequence for f about z0 , so that an = 0 for n < m and am = 0. Then for z D (z0 , ), (z z0 )m f (z ) = am + am+1 (z z0 ) + + a1 (z z0 )m1 + a0 (z z0 )m + , hence dm1 [(z z0 )m f (z )] = (m 1)!a1 + (z z0 )g (z ) dz m1 where g has a removable singularity at z0 . The result follows. (e) Suppose f is analytic at z0 and has a zero of order k at z0 . Then f /f has a simple pole at z0 and Res(f /f, z0 ) = k . Proof. There exists > 0 and a zero-free analytic function g on D(z0 , ) such that f (z ) = (z z0 )k g (z ) for z D(z0 , ). Then f (z ) = k (z z0 )k1 g (z ) + (z z0 )k g (z ), and hence for z D (z0 , ), k g (z ) f (z ) = + . f (z ) z z0 g (z ) Since g /g is analytic on D(z0 , ), it follows that f /f has a simple pole at z0 and Res(f /f, z0 ) = k . We are now ready for the main result of this section.

4.2.3

Residue Theorem

Let f be analytic on \ S , where S is a subset of with no limit point in . In other words, f is analytic on except for isolated singularities. Then for any closed path (or cycle) in \ S such that is -homologous to 0, we have f (w) dw = 2i
wS

n(, w) Res(f, w).

Proof. Let S1 = {w S : n(, w) = 0}. Then S1 Q = C \ {z / : n(, z ) = 0}. Since is -homologous to 0, Q is a subset of . Furthermore, by (3.2.5), Q is closed and bounded. Since S has no limit point in , S1 has no limit points at all. Thus S1 is a nite set. Consequently, the sum that appears in the conclusion of the theorem is the nite sum obtained by summing over S1 . Let w1 , w2 , . . . , wk denote the distinct points of S1 . [If S1 is empty, we are nished by Cauchys theorem (3.3.1).] Choose positive numbers r1 , r2 , . . . , rk so small that D (wj , rj ) \ S,
k

j = 1, 2, . . . , k.

Let be the cycle j =1 n(, wj )j , where j is the positively oriented boundary of D(wj , rj ). Then is cycle in the open set \ S , and you can check that if z / \ S , then

Ch: 1

69 (4-8)

TOC

Index

4.2. RESIDUE THEORY n(, z ) = n(, z ). Since f is analytic on \ S , it follows from (3.3.7) that f (w) dw. But by denition of ,
k k

f (w) dw =

f (w) dw =
j =1

n(, wj )
j

f (w) dw = 2i
j =1

n(, wj ) Res(f, wj )

by part (a) of (4.2.2). In many applications of the residue theorem, the integral f (w) dw is computed by evaluating the sum 2i wS n(, w) Res(f, w). Thus it is important to have methods available for calculating residues. For example, (4.2.2d) is useful when f is a rational function, since the only singularities of f are poles. The residue theorem can also be applied to obtain a basic geometric property of analytic functions called the argument principle. Before discussing the general result, lets look at a simple special case. Suppose z traverses the unit circle once in the positive sense, that is, z = eit , 0 t 2 . Then the argument of z 2 , namely 2t, changes by 4 , so that z 2 makes two revolutions around the origin. Thus the number of times that z 2 winds around the origin as z traverses the unit circle is the number of zeros of z 2 inside the circle, counting multiplicity. The index of a point with respect to a closed path allows us to formalize the notion of the number of times that f (z ) winds around the origin as z traverses a path . For we are looking at the net number of revolutions about 0 of f ( (t)), a t b, and this, as we have seen, is n(f , 0). We may now state the general result.

4.2.4

Argument Principle

Let f be analytic on , and assume that f is not identically zero on any component of . If Z (f ) = {z : f (z ) = 0} and is any closed path in \ Z (f ) such that is -homologous to 0,then n(f , 0) = n(, z )m(f, z )
z Z (f )

where m(f, z ) is the order of the zero of f at z . Proof. The set S = Z (f ) and the function f /f satisfy the hypothesis of the residue theorem. Applying it, we get 1 2i f (z ) dz = f (z ) n(, z ) Res(f /f, z ).
z Z (f ) z Z (f )

But the left side equals n(f , 0) by (3.2.3), and the right side equals by (4.2.2e).

n(, z )m(f, z )

4.2.5

Remarks

Assuming that for each z Z (f ), n(, z ) = 1 or 0, the argument principle says that the net increase in the argument of f (z ) as z traverses in the positive direction is equal to the number of zeros of f inside (n(, z ) = 1) with multiplicities taken into account.

Ch: 1

70 (4-9)

TOC

Index

10

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

There is a useful generalization of (4.2.4) to meromorphic functions. A function f is meromorphic on if f is analytic on except possibly for poles. That is, there is a subset S with no limit points in such that f is analytic on \ S and f has a pole at each point of S . For example, any rational function is meromorphic on C. More generally, the quotient f /g of two analytic functions is meromorphic, provided g is not identically zero on any component of . (This follows from (2.4.8) and (4.1.5).) Conversely, every meromorphic function is a quotient of two analytic functions. (This is a much deeper result, which will be proved in a later chapter.)

4.2.6

Denition

For f meromorphic on , let Z (f ) denote the set of zeros of f , and P (f ) the set of poles of f . If z Z (f ) P (f ), let m(f, z ) be the order of the zero or pole of f at z .

4.2.7

Argument Principle for Meromorphic Functions

Suppose f is meromorphic on . Then for any closed path (or cycle) in \ (Z (f ) P (f )) such that is -homologous to 0, we have n(f , 0) = n(, z )m(f, z ) n(, z )m(f, z ).
z P (f )

z Z (f )

Under certain conditions, the argument principle allows a very useful comparison of the number of zeros of two functions.

Proof. Take S = Z (f ) P (f ), and apply the residue theorem to f /f . The analysis is the same as in the proof of (4.2.4), if we note that if z0 P (f ), then Res(f /f, z0 ) = m(f, z0 ). To see this, write f (z ) = g (z )/(z z0 )k where k = m(f, z0 ) and g is analytic at z0 , with g (z0 ) = 0. Then f (z )/f (z ) = [g (z )/g (z )] [k/(z z0 )].

4.2.8

Rouch es Theorem

Suppose f and g are analytic on , with neither f nor g identically zero on any component of . Let be a closed path in such that is -homologous to 0. If |f (z ) + g (z )| < |f (z )| + |g (z )| for each z , then n(, z )m(f, z ) =
z Z (f ) z Z (g )

(1)

n(, z )m(g, z ).

Thus f and g have the same number of zeros, counting multiplicity and index. Proof. The inequality (1) implies that \ [Z (f ) P (f )], and hence by the argument principle, applied to each of f and g , we obtain n(f , 0) = n(, z )m(f, z )
z Z (f )

and n(g , 0) =

n(, z )m(g, z ).
z Z (g )

Ch: 1

71 (4-10)

TOC

Index

4.2. RESIDUE THEORY

11

But again by (1), |f ( (t)) + g ( (t))| < |f ( (t))| + |g ( (t))| for all t in the domain of the closed path . Therefore by the generalized dog-walking theorem (Problem 3.2.4), n(f , 0) = n(g , 0). The result follows.

4.2.9

Remarks

Rouch es theorem is true for cycles as well. To see this, suppose that is the formal sum k1 1 + + kr r . Then just as in the proof of (4.2.8), we have n(f , 0) = z Z (g ) n(, z )m(g, z ). But now |f (z ) + g (z )| < z Z (f ) n(, z )m(f, z ) and n(g , 0) = implies, as before, that n(f j , 0) = n(g j , 0) |f (z )| + |g (z )| for each z = r j =1 j for j = 1, . . . , r, hence n(f , 0) = n(g , 0) and the proof is complete. In the hypothesis of (4.2.8), (1) is often replaced by |f (z ) g (z )| < |f (z )| for each z . (2)

But now if (2) holds, then |f (z ) + (g (z ))| < |f (z )| |f (z )| + | g (z )| on , so f and g , hence f and g , have the same number of zeros.

Problems
1. Let f (z ) = (z 1)(z 3 + 4i)/(z + 2i)2 , and let be as shown in Figure 4.2.1. Find n(f , 0), and interpret the result geometrically.

-2i

. i .3-4i

Figure 4.2.1 2. Use the argument principle to nd (geometrically) the number of zeros of z 3 z 2 +3z +5 in the right half plane. 3. Use Rouch es theorem to prove that any polynomial of degree n 1 has exactly n zeros, counting multiplicity. 4. Evaluate the following integrals using residue theory or Cauchys theorem. x sin ax (b) (x2 +1)(x (a) xx 4 +4 dx, a > 0 2 +2x+2) dx (c) (f)
1 (x2 4x+5)2 cos x 0 x2 +1

dx (d)

2 cos 0 5+4 cos

(e)

1 0 x4 +a4

dx, a > 0

dx (g)

2 (sin )2n 0

Ch: 1

72 (4-11)

TOC

Index

12

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

y -5+10i 10+10i

-5+5i

5+5i

x -5-21 -5-5i
Figure 4.2.2 5. Evaluate
Log z 1+ez

10-2i 5-5i

dz along the path indicated in Figure 4.2.2. (c) z cos(1/z ).

z

6. Find the residue at z = 0 of (a) csc2 z , (b) z 3 csc(z 2 ),

7. Find the residue of sin(e /z ) at z = 0. (Leave the answer in the form of an innite series.) 8. The results of this exercise are necessary for the calculations that are to be done in Problem 9. (a) Show that for any r > 0,
/2 0

er sin d

(1 er ). 2r

(Hint: sin 2/ for 0 /2.) (b) Suppose f has a simple pole at z0 , and let be a circular arc with center z0 and radius which subtends an angle at z0 , 0 < 2 (see Figure 4.2.3). Prove that
0

lim

f (z ) dz = i Res(f, z0 ).

In particular, if the are semicircular arcs ( = ), then

0

lim

f (z ) dz = i Res(f, z0 ) = (1/2)2i Res(f, z0 ).

9. (a) Show that in Figure 4.2.4.

(Hint: f (z ) [Res(f, z0 )/(z z0 )] has a removable singularity at z0 .)

sin x x

dx = by integrating eiz /z on the closed path R, indicated

Ch: 1

73 (4-12)

TOC

Index

4.2. RESIDUE THEORY

13

zo

Figure 4.2.3
1 (b) Show that 0 cos x2 dx = 0 sin x2 dx = 2 /2. (Integrate eiz around the 2 closed path indicated in Figure 4.2.5; assume as known the result that 0 ex dx = 1 2 .) Log(z +i) x +1) (c) Compute 0 ln( around the closed path of Figure x2 +1 dx by integrating z 2 +1 4.2.6. /2 /2 (d) Derive formulas for 0 ln cos d and 0 ln sin d by making the change of variable x = tan in (c).
2

Figure 4.2.4

/4

Figure 4.2.5 10. Use Rouch es theorem to show that all the zeros of z 4 + 6z + 3 are in |z | < 2, and three of them are in 1 < |z | < 2.

11. Suppose f is analytic on an open set D(0, 1), and |f (z )| < 1 for |z | = 1. Show that for each n, the function f (z ) z n has exactly n zeros in D(0, 1), counting multiplicity. In particular, f has exactly one xed point in D(0, 1).

12. Prove the following version of Rouch es theorem. Suppose K is compact, is an open subset of K , f and g are continuous on K and analytic on , and we have the

Ch: 1

74 (4-13)

TOC

Index

14

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

Figure 4.2.6 inequality |f (z ) + g (z )| < |f (z )| + |g (z )| for every z K \ . Show that f and g have the same number of zeros in , that is, m(f, z ) =
z Z (f ) z Z (g )

m(g, z ).

[Hint: Note that Z (f ) Z (g ) {z : |f (z ) + g (z )| = |f (z )| + |g (z )|, and the latter set is a compact subset of . Now apply the hexagon lemma and (4.2.9).] 13. Show that
1 eiux 1+x2

dx = e|u| for real u.

14. Evaluate the integral of exp[sin(1/z )] around the unit circle |z | = 1.

15. Suppose f and g are analytic at z0 . Establish the following: (a) If f has a zero of order k and g has a zero of order k + 1 at z0 , then f /g has a simple pole at z0 and Res(f /g, z0 ) = (k + 1)f (k) (z0 )/g (k+1) (z0 ). (The case k = 0 is allowed.) (b) If f (z0 ) = 0 and g has a zero of order 2 at z0 , then f /g has a pole of order 2 at z0 and Res(f /g, z0 ) = 2 2 f (z0 )g (z0 ) f (z0 ) . g (z0 ) 3 [g (z0 )]2

18. Evaluate

17. Let f be analytic on D(0, 1) with f (0) = 0. Suppose > 0, 0 < r < 1, and min|z|=r |f (z )| . Prove that D(0, ) f (D(0, r)). ez 1 1 + cos + z +1 z e dz.

16. Show that the equation 3z = ez has exactly one root in |z | < 1.

C (1+i,2)

z2

19. Suppose that P and Q are polynomials, the degree of Q exceeds that of P by at least 2, and the rational function P/Q has no poles on the real axis. Prove that [P (x)/Q(x)] dx is 2i times the sum of the residues of P/Q at its poles in the upper half plane. Then compute this integral with P (x) = x2 and Q(x) = 1 + x4 .

Ch: 1

75 (4-14)

TOC

Index

4.3. THE OPEN MAPPING THEOREM FOR ANALYTIC FUNCTIONS

15

20. Prove that the equation ez 3z 7 = 0 has seven roots in the unit disk |z | < 1. More generally, if |a| > e and n is a positive integer, prove that ez az n has exactly n roots in |z | < 1. 22. Show that f (z ) = z 7 5z 4 + z 2 2 has exactly 4 zeros inside the unit circle. 24. Show that all the roots of z 5 + z + 1 = 0 satisfy |z | < 5/4. 21. Prove that ez = 2z + 1 for exactly one z D(0, 1).

23. If f (z ) = z 5 + 15z + 1, prove that all zeros of f are in {z : |z | < 2}, but only one zero of f is in {z : |z | < 1/2}.

25. Let {fn } be a sequence of analytic functions on an open connected set such that fn f uniformly on compact subsets of . Assume that f is not identically zero, and let z0 . Prove that f (z0 ) = 0 i there is a subsequence {fnk } and a sequence {zk } such that zk z0 and fnk (zk ) = 0 for all k . (Suggestion: Rouch es theorem.) 26. Let p(z ) = an z n + + a1 z + a0 , an = 0, dene q (z ) = a0 z n + + an1 z + an , and put f (z ) = a0 p(z ) an q (z ). Assume that p has k 0 zeros in |z | < 1, but no zeros on |z | = 1. Establish the following. (a) For z = 0, q (z ) = z n p(1/z ). (b) q has n k zeros in |z | < 1. (c) |p(z )| = |q (z )| for |z | = 1. (d) If |a0 | > |an |, then f also has k zeros in |z | < 1, while if |a0 | < |an |, then f has n k zeros in |z | < 1. (e) If |a0 | > |an |, then p has at least one zero in |z | > 1, while if |a0 | < |an |, then p has at least one zero in |z | < 1.

4.3

The Open Mapping Theorem for Analytic Functions

Our aim in this section is to show that a non-constant analytic function on a region maps to a region, and that a one-to-one analytic function has an analytic inverse. These facts, among others, are contained in the following theorem.

4.3.1

Open Mapping Theorem

(b) Let be the positively oriented boundary of D(z0 , ), let W0 be the component of C \ (f ) that contains w0 , and let 1 = D(z0 , ) f 1 (W0 ). Then f is a k -to-one map of 1 \ {z0 } onto W0 \ {w0 }.

(a) There exists > 0 such that D(z0 , ) and such that neither f w0 nor f has a zero in D(z0 , ) \ {z0 }.

Let f be a non-constant analytic function on an open connected set . Let z0 and w0 = f (z0 ), and let k = m(f w0 , z0 ) be the order of the zero which f w0 has at z0 .

Ch: 1

76 (4-15)

TOC

Index

16

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

(c) f is a one-to-one map of 1 onto W0 i f (z0 ) = 0. (d) f () is open. (e) f : C maps open subsets of onto open sets. (f) If f is one-to-one, then f 1 is analytic. Proof. (a) This follows from the identity theorem; the zeros of a non-constant analytic function and its derivative have no limit point in . (b) If w W0 , then by the argument principle, n(f ) is the number of zeros of f w in D(z0 , ). But n(f , w) = n(f , w0 ), because the index is constant on components of the complement of (f ) . Since n(f , w0 ) = k , and f has no zeros in D (z0 , ), it follows that for w = w0 , f w has exactly k zeros in D(z0 , ), all simple. This proves (b). (c) If f (z0 ) = 0, then k = 1. Conversely, if f (z0 ) = 0, then k > 1. (d) This is a consequence of (a) and (b), as they show that f (z0 ) is an interior point of the range of f . (e) This is a consequence of (d) as applied to an arbitrary open subdisk of . (f) Assume that f is one-to-one from onto f (). Since f maps open subsets of onto open subsets of f (), f 1 is continuous on f (). By (c), f has no zeros in , and Theorem 1.3.2 then implies that f 1 is analytic.

4.3.2

Remarks

If is not assumed to be connected, but f is non-constant on each component of , then the conclusions of (4.3.1) are again true. In particular, if f is one-to-one, then surely f is non-constant on components of and hence f 1 is analytic on f (). Finally, note that the maximum principle is an immediate consequence of the open mapping theorem. (Use (4.3.1d), along with the observation that given any disk D(w0 , r), there exists w D(w0 , r) with |w| > |w0 |.) The last result of this section is an integral representation theorem for f 1 in terms of the given function f . It can also be used to give an alternative proof that f 1 is analytic.

4.3.3

Theorem

Let f and g be analytic on and assume that f is one-to-one. Then for each z0 and each r such that D(z0 , r) , we have g (f 1 (w)) = 1 2i g (z )
C (z0 ,r )

f (z ) dw f (z ) w

for every w f (D(z0 , r)). In particular, with g (z ) = z , we have f 1 (w) = 1 2i z

C (z0 ,r )

f (z ) dw. f (z ) w

Ch: 1

77 (4-16)

TOC

Index

4.4. LINEAR FRACTIONAL TRANSFORMATIONS

17

f (z ) 1 Proof. Let w f (D(z0 , r)). The function h(z ) = g (z ) f ( (w)}, z )w is analytic on \ {f and hence by the residue theorem [or even (4.2.2a)],

1 2i

g (z )
C (z0 ,r )

f (z ) dz = Res(h, w). f (z ) w

But g is analytic at f 1 (w) and f w has a simple zero at f 1 (w) (because f is one-toone), hence (Problem 1) Res(h, w) = g (f 1 (w)) Res( f , w) = g (f 1 (w)) by (4.2.2e). f w

In Problem 2, the reader is asked to use the above formula to give another proof that f 1 is analytic on f ().

Problems
1. Suppose g is analytic at z0 and f has a simple pole at z0 . show that Res(gf, z0 ) = g (z0 ) Res(f, z0 ). Show also that the result is false if the word simple is omitted. 2. Let f be as in Theorem 4.3.3. Use the formula for f 1 derived therein to show that f 1 is analytic on f (). (Show that f 1 is representable in f () by power series.) 3. The goal of this problem is an open mapping theorem for meromorphic functions. Recall from (4.2.5) that f is meromorphic on if f is analytic on \ P where P is a subset of with no limit point in such that f has a pole at each point of P . Dene f (z ) = if z P , so that by (4.1.5b), f is a continuous map of into the extended . Prove that if f is non-constant on each component of , then f () is open plane C in C. 4. Suppose f is analytic on , D(z0 , r) , and f has no zeros on C (z0 , r). Let a1 , a2 , . . . , an be the zeros of f in D(z0 , r). Prove that for any g that is analytic on , 1 2i f (z ) g (z ) dz = f (z )
n

m(f, aj )g (aj )
j =1

C (z0 ,r)

where (as before) m(f, aj ) is the order of the zero of f at aj . 5. Let f be a non-constant analytic function on an open connected set . How does the open mapping theorem imply that neither |f | nor Re f nor Im f takes on a local maximum in ?

4.4

Linear Fractional Transformations

In this section we will study the mapping properties of a very special class of functions on C, the linear fractional transformations (also known as M obius transformations ).

Ch: 1

78 (4-17)

TOC

Index

18

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

4.4.1

Denition

Note that the condition ad bc = 0 guarantees that T is not constant. Also, if c = 0, then a = 0 and d = 0, so that the usual agreements regarding can be made, that is, T () = a/c if c = 0, if c = 0 and T (d/c) = if c = 0.

If a, b, c, d are complex numbers such that ad bc = 0, the linear fractional transformation C associated with a, b, c, d is dened by T :C az +b cz+d , z = , z = d/c T (z ) = a/c, z= , z = d/c.

onto C . Moreover, It follows from the denition that T is a one-to-one continuous map of C T is analytic on C \ {d/c} with a simple pole at the point d/c. Also, each such T is a composition of maps of the form (i) z z + B (translation) (ii) z z , where || = 1 (rotation) (iii) z z, > 0 (dilation) (iv) z 1/z (inversion).

To see that T is always such a composition, recall that if c = 0, then a = 0 = d, so T (z ) = |a/d| a/d b z+ , |a/d| d

and if c = 0, then T (z ) = (bc ad)/c2 a + . z + (d/c) c

Linear fractional transformations have the important property of mapping the family of lines and circles in C onto itself. This is most easily seen by using complex forms of equations for lines and circles.

4.4.2

Theorem

Let L = {z : zz + z + z + = 0} where and are real numbers, is complex, and s2 = > 0. If = 0, then L is a circle, while if = 0, then L is a line. Conversely, each line or circle can be expressed as one of the sets L for appropriate , , . Proof. First let us suppose that = 0. Then the equation dening L is equivalent to |z + (/)|2 = ( )/2 , which is the equation of a circle with center at / and radius s/||. Conversely, the circle with center z0 and radius r > 0 has equation |z z0 |2 = r2 , which is equivalent to zz z 0 z z0 z + |z0 |2 r2 = 0. This has the required form with = 1, = z0 , = |z0 |2 r2 . On the other hand, if = 0, then = 0, and

Ch: 1

79 (4-18)

TOC

Index

4.5. CONFORMAL MAPPING

19

the equation dening L becomes z + z + = 0, which is equivalent to Re(z )+ /2 = 0. This has the form Ax + By + /2 = 0 where z = x + iy and = A + iB , showing that L is a line in this case. Conversely, an equation of the form Ax + By + C = 0, where A and B are not both zero, can be written in complex form as Re(z ) + /2 = 0, where = A + iB and = 2C .

4.4.3

Theorem

Suppose L is a line or circle, and T is a linear fractional transformation. Then T (L) is a line or circle. Proof. Since T is a composition of maps of the types (i)-(iv) of (4.4.1), it is sucient to show that T (L) is a line or circle if T is any one of these four types. Now translations, dilations, and rotations surely map lines to lines and circles to circles, so it is only necessary to look at the case where T (z ) = 1/z . But if z satises zz + z + z + = 0, then w = 1/z satises ww + w + w + = 0, which is also an equation of a line or circle.

Note, for example, that if T (z ) = 1/z , = 0 and = 0, then L is a circle through the origin, but T (L), with equation w + w + = 0, is a line not through the origin. This is to be expected because inversion interchanges 0 and .

Linear fractional transformations also have an angle-preserving property that is possessed, more generally, by all analytic functions with non-vanishing derivatives. This will be discussed in the next section. Problems on linear fractional transformations will be postponed until the end of Section 4.5.

4.5

Conformal Mapping

We saw in the open mapping theorem that if f (z0 ) = 0, then f maps small neighborhoods of z0 onto neighborhoods of f (z0 ) in a one-to-one fashion. In particular, f maps smooth arcs (that is, continuously dierentiable arcs) through z0 onto smooth arcs through f (z0 ). Our objective now is to show that f preserves angles between any two such arcs. This is made precise as follows.

4.5.1

Denition

Suppose f is a complex function dened on a neighborhood of z0 , with f (z ) = f (z0 ) for all z near z0 but not equal to z0 . If there exists a unimodular complex number ei such that for all , f (z0 + rei ) f (z0 ) ei ei |f (z0 + rei ) f (z0 )| as r 0+ , then we say that f preserves angles at z0 . To gain some insight and intuitive feeling for the meaning of the above condition, f (z0 +rei )f (z0 ) note that for any and small r0 > 0, |f is a unit vector from f (z0 ) to (z0 +r0 ei )f (z0 )| f (z0 + r0 ei ). The vectors from z0 to z0 + rei , 0 < r r0 , have argument , so the

Ch: 1

80 (4-19)

TOC

Index

20

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

condition states that f maps these vectors onto an arc from f (z0 ) whose unit tangent vector at f (z0 ) has argument + . Since is to be the same for all , f rotates all short vectors from z0 through the xed angle . Thus we see that f preserves angles between tangent vectors to smooth arcs through z0 .

4.5.2

Theorem

Suppose f is analytic at z0 . Then f preserves angles at z0 i f (z0 ) = 0. Proof. If f (z0 ) = 0, then for any , lim f (z0 + rei ) f (z0 ) [f (z0 + rei ) f (z0 )]/rei f (z0 ) = ei lim+ = ei i i |f (z0 + re ) f (z0 )| |[f (z0 + re ) f (z0 )|]/r |f (z0 )|. r 0

r 0+

Thus the required unimodular complex number of Denition 4.5.1 is f (z0 )/|f (z0 )|. Conversely, suppose that f (z0 ) = 0. Assuming that f is not constant, f f (z0 ) has a zero of some order m > 1 at z0 , hence we may write f (z ) f (z0 ) = (z z0 )m g (z ) where g is analytic at z0 and g (z0 ) = 0. For any and small r > 0, f (z0 + rei ) f (z0 ) g (z0 + rei ) g (z0 + rei ) = eim = ei ei(m1) i i |f (z0 + re ) f (z0 )| |g (z0 + re )| |g (z0 + rei )| and the expression on the right side approaches ei ei(m1) g (z0 )/|g (z0 )| as r 0+ . Since the factor ei(m1) g (z0 )/|g (z0 )| depends on , f does not preserve angles at z0 . Indeed, the preceding shows that angles are increased by a factor of m, the order of the zero of f f (z0 ) at z0 .

A function f on that is analytic and has a nonvanishing derivative will be called a conformal map ; it is locally one-to-one and preserves angles. Examples are the exponential function and the linear fractional transformation (on their domains of analyticity). The angle-preserving property of the exponential function was illustrated in part (i) of (2.3.1), where it was shown that exp maps any pair of vertical and horizontal lines onto, respectively, a circle with center 0 and an open ray emanating from 0. Thus the exponential function preserves the orthogonality of vertical and horizontal lines.

Problems
1. Show that the inverse of a linear fractional transformation and the composition of two linear fractional transformations is again a linear fractional transformation. 2. Consider the linear fractional transformation T (z ) = (1 + z )/(1 z ). (a) Find a formula for the inverse of T . (b) Show that T maps |z | < 1 onto Re z > 0, |z | = 1 onto {z : Re z = 0} {}, and |z | > 1 onto Re z < 0. 3. Find linear fractional transformations that map (a) 1, i, 1 to 1, 0, 1 respectively. (b) 1, i, 1 to 1, i, 1 respectively.

Ch: 1

81 (4-20)

TOC

Index

4.6. ANALYTIC MAPPINGS OF ONE DISK TO ANOTHER

21

4. Let (z1 , z2 , z3 ) be a triple of distinct complex numbers. (a) Prove that there is a unique linear fractional transformation T with the property that T (z1 ) = 0, T (z2 ) = 1, T (z3 ) = . (b) Prove that if one of z1 , z2 , z3 is , then the statement of (a) remains true. (c) Let each of (z1 , z2 , z3 ) and (w1 , w2 , w3 ) be triples of distinct complex numbers ). Show that there is a unique linear fractional (or extended complex numbers in C transformation such that T (zj ) = wj , j = 1, 2, 3. 5. Let f be meromorphic on C and assume that f is one-to-one. Show that f is a linear fractional transformation. In particular, if f is entire, then f is linear, that is, a rst degree polynomial in z . Here is a suggested outline: (a) f has at most one pole in C, consequently is an isolated singularity of f . . (b) f (D(0, 1)) and f (C \ D(0, 1)) are disjoint open sets in C . (c) f has a pole or removable singularity at , so f is meromorphic on C (d) f has exactly one pole in C. (e) Let the pole of f be at z0 . If z0 = , then f is a polynomial, which must be of degree 1. If z0 C, consider g (z ) = 1/f (z ), z = z0 ; g (z0 ) = 0. Then g is analytic at z0 and g (z0 ) = 0. (f) f has a simple pole at z0 . (g) f (z ) [Res(f, z0 ))/(z z0 )] is constant, hence f is a linear fractional transformation.

4.6

Analytic Mappings of One Disk to Another

In this section we will investigate the behavior of analytic functions that map one disk into another. The linear fractional transformations are examples which are, in addition, one-to-one. Schwarzs lemma (2.4.16) is an important illustration of the type of conclusion that can be drawn about such functions, and will be generalized in this section. We will concentrate on the special case of maps of the unit disk D = D(0, 1) into itself. The following lemma supplies us with an important class of examples.

4.6.1

Lemma
za , 1 az

by Fix a D, and dene a function a on C a (z ) =

where the usual conventions regarding are made: a () = 1/a and a (1/a) = . into C whose inverse is a . Also, a is Then a is a one-to-one continuous map of C analytic on C \ {1/a} with a simple pole at 1/a (and a zero of order 1 at a). Thus a is analytic on a neighborhood of the closed disk D. Finally, a (D) = D, hence a (a) = 1 1 |a|2
2 and a (0) = 1 |a| .

a (D) = D,

a (z ) =

1 |a|2 (1 az )2

Ch: 1

82 (4-21)

TOC

Index

22

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

Proof. Most of the statements follow from the denition of a and the fact that it is a linear fractional transformation. To see that a maps |z | = 1 into itself, we compute, for |z | = 1, za za za = = = 1. 1 az z (1 az ) za
1 Thus by the maximum principle, a maps D into D. Since a = a (a computation shows that a (a (z )) = z ), and |a| < 1 i | a| < 1, it follows that a maps D onto D and maps D onto D. The formulas involving the derivative of a are veried by a direct calculation.

4.6.2

Remark

The functions a are useful in factoring out the zeros of a function g on D, because g (z ) and a (g (z )) have the same maximum modulus on D, unlike g (z ) and (z a)g (z ). In fact, if g is dened on the closed disk D, then za g (z ) = |g (z )| for |z | = 1. 1 az This property of the functions a will be applied several times in this section and the problems following it. We turn now to what is often called Picks generalization of Schwarzs lemma.

4.6.3

Theorem

Let f : D D be analytic. then for any a D and any z D, 1 f (a)f (z ) and |f (a)| 1 |f (a)|2 . 1 |a|2 (ii) f (z ) f (a) za 1 az (i)

Furthermore, if equality holds in (i) for some z = a, or if equality holds in (ii), then f is a linear fractional transformation. In fact, there is a unimodular complex number such 1 that with b = f (a), f is the composition b a = b a . That is, f (z ) = a (z ) + b , 1 + ba (z ) |z | < 1.

Proof. Let a D and set b = f (a). We are going to apply Schwarzs lemma (2.4.16) to the function g = b f a . First, since f maps D into D, so does g . Also, g (0) = b (f (a (0))) = b (f (a)) = b (b) = 0.

Ch: 1

83 (4-22)

TOC

Index

4.6. ANALYTIC MAPPINGS OF ONE DISK TO ANOTHER

23

By Schwarzs lemma, |g (w)| |w| for |w| < 1, and replacing w by a (z ) and noting that g (a (z )) = b (f (z )), we obtain (i). Also by (2.4.16), we have |g (0) 1. But by (4.6.1),
g (0) = b (f (a (0)))f (a (0))a (0) 2 = b (f (a))f (a)(1 |a| ) 1 f (a)(1 |a|2 ). = 1 |f (a)|2 )

Now if equality holds in (i) for some z = a, then |g (a (z ))| = |a (z )| for hence |g (w)| = |w| for some w = 0. If equality holds in (ii), then |g (0)| = case, (2.4.16) yields a unimodular complex number such that g (w) = w Set w = a (z ) to obtain b (f (z )) = a (z ), that is, f (z ) = b (a (z )) for

Thus the condition |g (0)| 1 implies (ii).

An important application of Theorem 4.6.3 is in characterizing the one-to-one analytic maps of D onto itself as having the form a where || = 1 and a D.

some z = a, 1. In either for |w| < 1. |z | < 1.

4.6.4

Theorem

Suppose f is a one-to-one analytic map of D onto D. then f = a for some unimodular and a D. Proof. Let a D be such that f (a) = 0 and let g = f 1 , so g (0) = a. Now since g (f (z )) = z , we have 1 = g (f (z ))f (z ), in particular, 1 = g (f (a))f (a) = g (0)f (a). Next, (4.6.3ii) implies that |g (0)| 1 |a|2 and |f (a)| 1/(1 |a|2 ). Thus 1 = |g (0)||f (a)| 1 |a|2 = 1. 1 |a|2

Necessarily then, |f (a)| = 1/(1 |a|2 ) (and |g (0)| = 1 |a|2 ). Consequently, by the condition for equality in (4.6.3ii), f = a , as required.

4.6.5

Remark

One implication of the previous theorem is that any one-to-one analytic map of D onto D actually extends to a homeomorphism of D onto D. We will see when we study the Riemann mapping theorem in the next chapter that more generally, if f maps D onto a special type of region , then f again extends to a homeomorphism of D onto . Our nal result is a characterization of those continuous functions on D which are analytic on D and have constant modulus on the boundary |z | = 1. The technique mentioned in (4.6.2) will be used.

4.6.6

Theorem

Suppose f is continuous on D, analytic on D, and |f (z )| = 1 for |z | = 1. Then there is a unimodular , nitely many points a1 , . . . , an in D, and positive integers k1 , . . . , kn ,

Ch: 1

84 (4-23)

TOC

Index

24 such that

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

f (z ) =
j =1

z aj 1 aj z

kj

In other words, f is, to within a multiplicative constant, a nite product of functions of the type a . (If f is constant on D, the product is empty and we agree that it is identically 1 in this case.) Proof. First note that |f (z )| = 1 for |z | = 1 implies that f has at most nitely many zeros in D. If f has no zeros in D, then by the maximum and minimum principles, f is constant on D. Suppose then that f has its zeros at the points a1 , . . . , an with orders k1 , . . . , kn respectively. Put
n

g (z ) =
j =1

z aj 1 aj z

kj

z D.

Then f /g has only removable singularities in D, the analytic extension of f /g has no zeros in D, and |f /g | = 1 on D. Again by the maximum and minimum principles, f /g is constant on D \ {a1 , . . . , an }. Thus f = g with || = 1.

Problems
1. Derive the inequality (4.6.3ii) directly from (4.6.3i). 2. Let f be an analytic map of D(0, 1) into the right half plane {z : Re z > 0}. Show that 1 + |z | 1 |z | |f (0)| |f (z )| |f (0)|, 1 + |z | 1 |z | and |f (0)| 2| Re f (0)|. Hint: Apply Schwarzs lemma to T f , where T (w) = (w f (0))/(w + f (0)). 3. Show that if f is an analytic map of D(0, 1) into itself, and f has two or more xed points, then f (z ) = z for all z D(0, 1). 4. (a) Characterize the entire functions f such that |f (z )| = 1 for |z | = 1 [see (4.6.6)]. (b) Characterize the meromorphic functions f on C such that |f (z )| = 1 for |z | = 1. (Hint: If f has a pole of order k at a D(0, 1), then [(z a)/(1 az )]k f (z ) has a removable singularity at a.) 5. Suppose that in Theorem 4.6.3, the unit disk D is replaced by D(0, R) and D(0, M ). That is, suppose f : D(0, R) D(0, M ). How are the conclusions (i) and (ii) modied in this case? (Hint: Consider g (z ) = f (Rz )/M .) 6. Suppose f : D(0, 1) D(0, 1) is continuous and f is analytic on D(0, 1). Assume that f has zeros at z1 , . . . , zn of orders k1 , . . . , kn respectively. Show that
n

z D(0, 1),

|f (z )|

j =1

z zj 1 zj z

kj

Ch: 1

85 (4-24)

TOC

Index

4.7. THE POISSON INTEGRAL FORMULA AND ITS APPLICATIONS

25

Suppose equality holds for some z D(0, 1) with z = zj , j = 1, . . . , n. Find a formula for f (z ).

4.7

The Poisson Integral Formula and its Applications

Our aim in this section is to solve the Dirichlet problem for a disk, that is, to construct a solution of Laplaces equation in the disk subject to prescribed boundary values. The basic tool is the Poisson integral formula, which may be regarded as an analog of the Cauchy integral formula for harmonic functions. We will begin by extending Cauchys theorem and the Cauchy integral formula to functions continuous on a disk and analytic on its interior.

4.7.1

Theorem

Suppose f is continuous on D(0, 1) and analytic on D(0, 1). Then (i) C (0,1) f (w) dw = 0 and (ii) f (z ) =
1 2i f (w ) C (0,1) wz

Proof. For 0 < r < 1, C (0,r) f (w) dw = 0 by Cauchys theorem. For n = 1, 2, . . . , put n fn (z ) = f ( n+1 z ). Then fn is analytic on D(0, n+1 n ) and the sequence {fn } converges to f uniformly on C (0, 1) [by continuity of f on D(0, 1)]. Hence C (0,1) fn (w) dw f (w) dw = 0, we have (i). To f (w) dw. Since C (0,1) fn (w) dw = n+1 n C (0, n ) C (0,1)
n+1

dw for all z D(0, 1).

prove (ii), we apply (i) to the function g , where g (w) =

f (w)f (z ) , wz

f (z ),

w=z w = z.

Note that the same proof works with only minor modications if D(0, 1) is replaced by an arbitrary disk D(z0 , R).

4.7.2

Denition
1 |z |2 |eit z |2 eit + z ; eit z

For z D(0, 1), dene functions Pz and Qz on the real line R by Pz (t) = and Qz (t) =

Pz (t) is called the Poisson kernel and Qz (t) the Cauchy kernel. We have Re[Qz (t)] = Re (eit + z )(eit z ) 1 |z |2 + zeit zeit = Re = Pz (t). |eit z |2 |eit z |2 1 r2 1 r2 = Pr (t ). = i(t i 2 re | |e ) r|2

Note also that if z = rei , then Pz (t) = |eit

Ch: 1

86 (4-25)

TOC

Index

26

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

Since |ei(t) r|2 = 1 2r cos(t ) + r2 , we see that Pr (t ) =

Thus for 0 r < 1, Pr (x) is an even function of x. Note also that Pr (x) is positive and decreasing on [0, ]. After these preliminaries, we can establish the Poisson integral formula for the unit disk, which states that the value of an analytic function at a point inside the disk is a weighted average of its values on the boundary, the weights being given by the Poisson kernel. The precise statement is as follows.

1 r2 1 r2 = = Pr ( t). 1 2r cos(t ) + r2 1 2r cos( t) + r2

4.7.3

Poisson Integral Formula

1 2 1 2
2

Suppose f is continuous on D(0, 1) and analytic on D(0, 1). then for z D(0, 1) we have f (z ) = and therefore
2

Pz (t)f (eit ) dt
0

Re f (z ) = Proof. By Theorem 4.7.1(ii), f (0) =

2

Pz (t) Re f (eit ) dt.

0

1 2i

C (0,1)

1 f (w) dw = w 2

f (eit ) dt,
0

hence f (0) = (1/2 ) 0 P0 (t)f (eit ) dt because P0 (t) 1. This takes care of the case z = 0. If z = 0, then again by (4.7.1) we have f (z ) = 1 2i f (w) dw wz and 0 = 1 2i f (w) dw, w 1/z

C (0,1)

C (0,1)

/ D(0, 1). Subtracting the second equation from the second equation holding because 1/z the rst, we get f (z ) = 1 2i [
C (0,1) 2

1 = 2 = 1 2

[
0 2

eit eit eit

[
0 2

1 = 2 1 = 2

eit zeit + ]f (eit ) dt z 1 zeit

1 1 it ]eit f (eit ) dt z e 1/z

1 1 ]f (w) dw wz w 1/z

[
0 2 0

which proves the rst formula. Taking real parts, we obtain the second.

1 |z |2 f (eit ) dt |eit z |2

z eit + it ]f (eit ) dt z e z

Ch: 1

87 (4-26)

TOC

Index

4.7. THE POISSON INTEGRAL FORMULA AND ITS APPLICATIONS

27

4.7.4

Corollary
1 2 2 0

For |z | < 1,

Pz (t) dt = 1.

Proof. Take f 1 in (4.7.3).

Using the formulas just derived for the unit disk D(0, 1), we can obtain formulas for functions dened on arbitrary disks.

4.7.5

Poisson Integral Formula for Arbitrary Disks

Let f be continuous on D(z0 , R) and analytic on D(z0 , R). Then for z D(z0 , R), f (z ) = 1 2
2

P(zz0 )/R (t)f (z0 + Reit ) dt.

0

In polar form, if z = z0 + rei , then f (z0 + rei ) = 1 2

2 0

Pr/R ( t)f (z0 + Reit ) dt.

Proof. Dene g on D(0, 1) by g (w) = f (z0 + Rw). Then (4.7.3) applies to g , and we obtain g (w) = 1 2
2 0

Pw (t)g (eit ) dt, |w| < 1.

If z D(z0 , R), then w = (z z0 )/R D(0, 1) and f (z ) = g ( z z0 1 )= R 2

2

P(zz0 )/R (t)f (z0 + Reit ) dt

0

We now have the necessary machinery available to solve the Dirichlet problem for disks. Again, for notational reasons we will solve the problem for the unit disk D(0, 1). If desired, the statement and proof for an arbitrary disk can be obtained by the same technique we used to derive (4.7.5) from (4.7.3).

which establishes the rst formula. For the second, apply (4.7.2). [See the discussion beginning with Note also that . . . .]

4.7.6

The Dirichlet Problem

Suppose u0 is a real-valued continuous function on C (0, 1). Dene a function u on D(0, 1) by u(z ) = u0 (z )
1 2 2 0

Pz (t)u0 (e ) dt

it

for |z | = 1, for |z | < 1.

Ch: 1

88 (4-27)

TOC

Index

28

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

Then u is continuous on D(0, 1) and harmonic on D(0, 1). Furthermore (since Pz is the real part of Qz ), for z D(0, 1), u(z ) = Re 1 2
2

Qz (t)u0 (eit ) dt .
0

In particular, the given continuous function u0 on C (0, 1) has a continuous extension to D(0, 1) which is harmonic on the interior D(0, 1).
it Proof. The function z 21 0 Qz (t)u0 (e ) dt is analytic on D (0, 1) by (3.3.3), and therefore u is harmonic, hence continuous, on D(0, 1). All that remains is to show that u is continuous at points of the boundary C (0, 1). We will actually show that u(rei ) u0 (ei ) uniformly in as r 1. Since u0 is continuous on C (0, 1), this will prove that u is continuous at each of point of C (0, 1), by the triangle inequality. Thus let and r be real numbers with 0 < r < 1. Then by (4.7.2), (4.7.4) and the denition of u(z ), 2

u(rei ) u0 (ei ) =

1 2

2 0

Pr ( t)[u0 (eit ) u0 (ei )] dt.

Make the change of variable x = t and recall that Pr is an even function. The above integral becomes 1 2
2

Pr (x)[u0 (ei(+x) ) u0 (ei )] dx,

and the limits of integration can be changed to and , because the integrand has 2 as a period. Now x with 0 < < and write the last integral above as the sum, 1 2

1 2

1 2

We can estimate each of these integrals. The rst and third have absolute value at most 2 sup{|u0 (eit )| : t }Pr ( ), because Pr (x) is a positive and decreasing function on [0, ] and Pr (x) = Pr (x). The middle integral has absolute value at most sup{|u0 (ei(+x) ) u0 (ei )| : x }, by (4.7.4). But for xed > 0, Pr ( ) 0 as r 1, while sup{|u0 (ei(+x) ) u0 (ei )| : x } approaches 0 as 0, uniformly in because u0 is uniformly continuous on C (0, 1). Putting this all together, we see that given > 0 there is an r0 , 0 < r0 < 1, such that for r0 < r < 1 and all , we have |u(rei ) u0 (ei )| < . This, along with the continuity of u0 on C (0, 1), shows that u is continuous at each point of C (0, 1).

4.7.7

Uniqueness of Solutions to the Dirichlet Problem

We saw in (2.4.15) that harmonic functions satisfy the maximum and minimum principles. Specically, if u is continuous on D(0, 1) and harmonic on D(0, 1), then max u(z ) = max u(z ) and min u(z ) =
z D (0,1) z C (0,1) z D (0,1) z C (0,1)

min u(z ).

Ch: 1

89 (4-28)

TOC

Index

4.7. THE POISSON INTEGRAL FORMULA AND ITS APPLICATIONS

29

Thus if u 0 on C (0, 1), then u 0 on D(0, 1). Now suppose that u1 and u2 are solutions to a Dirichlet problem on D(0, 1) with boundary function u0 . Then u1 u2 is continuous on D(0, 1), harmonic on D(0, 1), and identically 0 on C (0, 1), hence identically 0 on D(0, 1). Therefore u1 u2 , so the solution to any given Dirichlet problem is unique. Here is a consequence of the uniqueness result.

4.7.8

Poisson Integral Formula for Harmonic Functions

Suppose u is continuous on D(0, 1) and harmonic on D(0, 1). Then for z D(0, 1), we have u(z ) = 1 2
2

Pz (t)u(eit ) dt.
0

More generally, if D(0, 1) is replaced by D(z0 , R), then u(z ) = equivalently, u(z0 + rei ) = 1 2
2 0

1 2

P(zz0 )/R (t)u(z0 + Reit ) dt;

0

Pr/R ( t)u(z0 + Reit ) dt

for 0 r < R and all . Proof. The result for D(0, 1) follows from (4.7.6) and (4.7.7). To prove the result for D(z0 , R), we apply (4.7.6) and (4.7.7) to u (w) = u(z0 + Rw), w D(0, 1). If z = z0 + rei , 0 r < R, then u(z ) = u ((z z0 )/R), hence u(z ) = 1 2
2

P(zz0 )/R (t)u (eit ) dt =

0

1 2

2 0

Pr/R ( t)u(z0 + Reit ) dt

as in (4.7.5). The Poisson integral formula allows us to derive a mean value property for harmonic functions.

4.7.9

Corollary
2

Suppose u is harmonic on an open set . If z0 and D(z0 , R) , then u(z0 ) = 1 2 u(z0 + Reit ) dt.
0

That is, u(z0 ) is the average of its values on circles with center at z0 . Proof. Apply (4.7.8) with r = 0. It is interesting that the mean value property characterizes harmonic functions.

Ch: 1

90 (4-29)

TOC

Index

30

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

4.7.10

Theorem

Suppose is a continuous, real-valued function on such that whenever D(z0 , R) , 2 it it is true that (z0 ) = 21 0 (z0 + Re ) dt. Then is harmonic on . Proof. Let D(z0 , R) be any disk such that D(z0 , R) . Let u0 be the restriction of to the circle C (z0 , R) and apply (4.7.6) [for the disk D(z0 , R)] to produce a continuous function u on D(z0 , R) such that u = u0 = on C (z0 , R). We will show that = u on D(z0 , R), thereby proving that is harmonic on D(z0 , R). Since D(z0 , R) is an arbitrary subdisk, this will prove that is harmonic on . The function u is continuous on D(z0 , R), and hence assumes its maximum and minimum at some points z1 and z2 respectively. If both z1 and z2 belong to C (z0 , R), then since u = on C (z0 , R), the maximum and minimum values of u are both 0. It follows that u 0 on D(z0 , R) and we are nished. On the other hand, suppose that (say) z1 belongs to the open disk D(z0 , R). Dene a set A by A = {z D(z0 , R) : ( u)(z ) = ( u)(z1 )}. Then A is closed in D(z0 , R) by continuity of u. We will also show that A is open, and thus conclude by connectedness that A = D(z0 , R). For suppose that a A and r > 0 is chosen so that D(a, r) D(z0 , R). Then for 0 < r we have (a) u(a) = 1 2
2 0

[(a + eit ) u(a + eit )] dt.

Since (a + eit ) u(a + eit ) (a) u(a), it follows from Lemma 2.4.11 that u is constant on D(a, r). Thus D(a, r) A, so A is open. A similar argument is used if z2 D(z0 , R).

Remark
The above proof shows that a continuous function with the mean value property that has an absolute maximum or minimum in a region is constant.

Problems
1. Let Qz (t) be as in (4.7.2). Prove that
1 2 2 0

Qz (t) dt = 1.

2. Use (4.7.8) to prove Harnacks inequality : Suppose u satises the hypothesis of (4.7.8), and in addition u 0. Then for 0 r < 1 and all , 1r 1+r u(0) u(rei ) u(0). 1+r 1r 3. Prove the following analog (for harmonic functions) of Theorem 2.2.17. Let {un } be a sequence of harmonic functions on such that un u uniformly on compact subsets of . Then u is harmonic on . (Hint: If D(z0 , R) , the Poisson integral formula holds for u on D(z0 , R).)

Ch: 1

91 (4-30)

TOC

Index

4.8. THE JENSEN AND POISSON-JENSEN FORMULAS

31

4. In Theorem 1.6.2 we showed that every harmonic function is locally the real part of an analytic function. Using results of this section, give a new proof of this fact. 5. Let be a bounded open set and a closed path such that the following conditions are satised: (a) = , the boundary of . (b) There exists z0 such that for every , 0 < 1, the path = z0 + ( z0 ) has its range in (see Figure 4.7.1). If f is continuous on and analytic on , show that f (w) dw = 0 and n(, z )f (z ) =

1 2i

f (w) dw, z . wz

Outline: (i) First show that must be starlike with star center z0 by showing that if z , then the ray [z0 , z, ) meets at some point . By (a) and (b), [z0 , ) . Next show that z [z0 , ), hence [z0 , z ] . (ii) The desired conclusions hold with replaced by ; let 1 to complete the proof. 6. (Poisson integral formula for a half plane). Let f be analytic on {z : Im z > 0} and continuous on {z : Im z 0}. If u = Re f , establish the formula u(x, y ) = 1

yu(t, 0) dt, Im z > 0 (t x)2 + y 2

under an appropriate hypothesis on the growth of f as z . (Consider the path indicated back in Figure 4.2.6. Write, for Im z > 0, f (z ) = (2i)1 [f (w)/(w z )] dw and 0 = (2i)1 [f (w)/(w z )] dw by using either Problem 5 or a technique similar to that given in the proof of (4.7.1). Then subtract the second equation from the rst.)

(t)

.z0

Figure 4.7.1

4.8

The Jensen and Poisson-Jensen Formulas

Suppose f is continuous on D(0, R), analytic on D(0, R) and f has no zeros in D(0, R). Then we know that f has an analytic logarithm on D(0, R) whose real part ln |f | is

Ch: 1

92 (4-31)

TOC

Index

32

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

continuous on D(0, R) and harmonic on D(0, R). Thus by (4.7.8), the Poisson integral formula for harmonic functions, we have ln |f (z )| = or in polar form, ln |f (rei )| = 1 2
2 0

1 2

2 0

Pz/R (t) ln |f (Reit )| dt

Pr/R ( t) ln |f (Reit )| dt.

If f has zeros in D(0, R), then this derivation fails. However, the above formula can be modied to take the zeros of f into account.

4.8.1

Poisson-Jensen Formula

Suppose that f is continuous on D(0, R), analytic on D(0, R) and that f has no zeros on C (0, R). Let a1 , . . . , an be the distinct zeros of f in D(0, R) with multiplicities k1 , . . . , kn respectively. Then for z D(0, R), z unequal to any of the aj , we have
n

ln |f (z )| =

kj ln
j =1

R(z aj ) 1 + 2 R aj z 2

2 0

Pz/R (t) ln |f (Reit | dt.

Since the product has modulus one when |z | = 1 we have |f (z )| = |g (z )| for |z | = 1. Thus if f (z ) = 0, then
n

Proof. We rst give a proof for the case R = 1. By (4.6.2), there is a continuous function g on D(0, 1), analytic on D(0, 1), such that g has no zeros in D(0, 1) and such that n kj z a j g (z ). f (z ) = 1 a z j j =1

ln |f (z )| =

kj ln
j =1

z aj + ln |g (z )|. 1 aj z

But g has no zeros in D(0, 1), so by the discussion in the opening paragraph of this section, ln |g (z )| = 1 2
2 0

Pz (t) ln |g (eit )| dt =

1 2

2 0

Pz (t) ln |f (eit )| dt.

This gives the result for R = 1. To obtain the formula for arbitrary R, we apply what was just proved to F (w) = f (Rw), |w| 1. Thus
n

ln |F (w)| =

kj ln
j =1

w (aj /R) 1 + 1 (aj w/R) 2

2 0

Pw (t) ln |F (eit )| dt.

Ch: 1

93 (4-32)

TOC

Index

4.8. THE JENSEN AND POISSON-JENSEN FORMULAS If we let z = Rw and observe that w (aj /R) R(z aj ) = 2 , 1 (aj w/R) R aj z we have the desired result. The Poisson-Jensen formula has several direct consequences.

33

4.8.2

Corollary
2

Assume that f satises the hypothesis of (4.8.1). Then

it (a) ln |f (z )| 21 0 Pz/R (t) ln |f (Re )| dt. If in addition, f (0) = 0, then

(b) ln |f (0)| =

(c) ln |f (0)| Part (b) is known as Jensens formula. Proof. It follows from (4.6.1) and the proof of (4.8.1) that R(z aj ) < 1, R2 aj z hence kj ln R(z aj ) < 0, R2 aj z

n 1 j =1 kj ln |aj /R| + 2 2 1 it 2 0 ln |f (Re )| dt.

2 0

ln |f (Reit )| dt, hence

proving (a). Part (b) follows from (4.8.1) with z = 0, and (c) follows from (b). Jensens formula (4.8.2b) does not apply when f (0) = 0, and the Poisson-Jensen formula (4.8.1) requires that f have no zeros on C (0, R). It is natural to ask whether any modications of our formulas are available so that these situations are covered. First, if f has a zero of order k at 0, with f (z ) = 0 for |z | = R, then the left side of Jensens formula is modied to k ln R + ln |f (k) (0)/k !| rather than ln |f (0)|. This can be veried by considering f (z )/z k and is left as Problem 1 at the end of the section. However, if f (z ) = 0 for some z C (0, R), then the situation is complicated for several reasons. For example, it is possible that f (z ) = 0 for innitely many points on C (0, R) without being identically zero on D(0, R) if f is merely assumed continuous on D(0, R) and analytic on D(0, R). Thus ln |f (z )| = at innitely many points in C (0, R) and so the Poisson integral of ln |f | does not a ` priori exist. It turns out that the integral does exist in the sense of Lebesgue, but Lebesgue integration is beyond the scope of this text. Thus we will be content with a version of the Poisson-Jensen formula requiring analyticity on D(0, R), but allowing zeros on the boundary.

4.8.3

Poisson-Jensen Formula, Second Version

Let f be analytic and not identically zero on D(0, R). Let a1 , . . . , an be the zeros of f in D(0, R), with multiplicities k1 , . . . , kn respectively. Then for z D(0, R) \ Z (f ),
n

ln |f (z )| =

kj ln
j =1

R(z aj ) 1 + R2 aj z 2

2 0

Pz/R (t) ln |f (Reit )| dt

Ch: 1

94 (4-33)

TOC

Index

34

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

where the integral exists as an improper Riemann integral. Proof. Suppose that in addition to a1 , . . . , an , f has zeros on C (0, R) at an+1 , . . . , am with multiplicities kn+1 , . . . , km . There is an analytic function g on D(0, R) with no zeros on C (0, R) such that f (z ) = (z an+1 )kn+1 (z am )km g (z ). (1)

The function g satises the hypothesis of (4.8.1) and has the same zeros as f in D(0, R). Now if z D(0, R) \ Z (f ), then
n

ln |f (z )| = But by applying (4.8.1) to g we get

n

j =1

kj ln |z aj | + ln |g (z )|.

ln |g (z )| =

ln
j =1

R(z aj ) 1 + R2 aj z 2

2 0

Pz/R (t) ln |g (Reit )| dt,

so the problem reduces to showing that

m j =n+1

kj ln |z aj | +

1 2

2 0

Pz/R (t) ln |g (Reit )| dt =

1 2

2 0

Pz/R (t) ln |f (Reit )| dt.

Since by (1), f (Reit ) = [ to show that

n it j =1 (Re

aj )kj ]g (Reit ), 0 t 2 , we see that it is sucient 1 2

2 0

ln |z aj | =

Pz/R (t) ln |Reit aj | dt

for j = n + 1, . . . , m. In other words, the Poisson integral formula (4.7.8) holds for the functions u(z ) = ln |z a| when |a| = R (as well as for |a| < R. This is essentially the content of the following lemma, where to simplify the notation we have taken R = 1 and a=1

4.8.4

Lemma
2 0

For |z | < 1, ln |z 1| = 1 2 Pz (t) ln |eit 1| dt,

where the integral is to be understood as an improper Riemann integral at 0 and 2 . In particular, 1 2

2 0

ln |eit 1| dt = 0.

Ch: 1

95 (4-34)

TOC

Index

4.8. THE JENSEN AND POISSON-JENSEN FORMULAS

35

Proof. We note rst that the above improper integral exists, because if 0 t , then |eit 1| = 2(1 cos t) = 2 sin(t/2) 2t/ . Therefore Pz (t) ln |eit 1| Pz (t) ln(2t/ ) = Pz (t)[ln(2/ ) + ln t]. Since the improper integral 0 ln t dt exists by elementary calculus and Pz (t) is continuous, the above inequalities imply that lim+ 1 2
/2

Pz (t) ln |eit 1| dt > and lim+

0

1 2

Pz (t) ln |eit 1| dt > .

Thus it remains to show that the value of the improper Riemann integral in the statement of the lemma is ln |z 1|. We will use a limit argument to evaluate the integral I= For r > 1, dene Ir = 1 2
2 0

1 2

2 0

Pz (t) ln |eit 1| dt.

Pz (t) ln |eit r| dt.

We will show that Ir I as r 1+ . Now for any xed r > 1, the function z ln |z r| is continuous on D(0, 1) and harmonic in D(0, 1), hence by (4.7.8), ln |z r| = Ir . Since ln |z r| ln |z 1| as r 1+ , this will show that I = ln |z 1|, completing the proof. So consider, for r > 1, |Ir I | = 1 2
2

Pz (t) ln
0

eit r 1 dt = it e 1 2

Pz (t) ln
0

eit r dt. eit 1

(The outer absolute values may be removed because |eit r| > |eit 1| and therefore the integrand is positive.) Using the 2 -periodicity of the integrand, we may write
2 0

=
0

+
0

and since Pz (t) = Pz (t), this becomes

+
0 0

=2
0

Now if 0 t , then eit 1 + 1 r 1r eit r = = 1 + it . it e 1 eit 1 e 1

But as we noted at the beginning of the proof, |eit 1| 2t/ , so the above expression is bounded in absolute value by 1 + [ (r 1)/2t]. Thus 0 < ln eit r (r 1) . ln 1 + eit 1 2t

Ch: 1

96 (4-35)

TOC

Index

36

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

Also, the Poisson kernel satises Pz (t) = 1 |z |2 (1 |z |)(1 + |z |) 1 + |z | = , it 2 2 |e z | (1 |z |) 1 |z |

an estimate that was used to establish Harnacks inequality. (See the solution to Section 4.7, Problem 2. Thus we now have 0 < |Ir I | Now x , 0 < < , and write

1 + |z | 1 1 |z |

ln 1 +
0

(r 1) dt. 2t

ln 1 +
0

(r 1) dt = 2t

ln 1 +
0

(r 1) dt + 2t

ln 1 +

(r 1) dt. 2t

Since the integral on the left side is nite (this is essentially the same as saying that ln t dt > ), and the integrand increases as r increases (r > 1), the rst integral on 0 the right side approaches 0 as 0+ , uniformly in r. On the other hand, the second integral on the right side is bounded by ( ) ln(1 + [ (r 1)/2 ]), which for xed > 0, approaches 0 as r 1+ . This completes the proof of the lemma, and as we noted earlier, nishes the proof of (4.8.3). The Poisson-Jensen formula has a number of interesting corollaries, some of which will be stated below. The proof of the next result (4.8.5), as well as other consequences, will be left for the problems.

4.8.5

Jensens Formula, General Case

Let f be analytic on an open disc D(0, R) and assume that f 0. Assume that f has a zero of order k 0 at 0 and a1 , a2 , . . . are the zeros of f in D(0, R) \ {0}, each appearing as often as its multiplicity and arranged so that 0 < |a1 | |a2 | . Then for 0 < r < R we have k ln r + ln f (k) (0) = k!
n( r )

ln
j =1

aj 1 + r 2

2 0

ln |f (reit )| dt

where n(r) is the number of terms of the sequence a1 , a2 , . . . that are in the disk D(0, r).

Problems
1. Prove (4.8.5). 2. Let f be as in (4.8.2), except that instead of being analytic on all of D(0, R), f has poles at b1 , . . . , bm in D(0, R) \ {0}, of orders l1 , . . . , lm respectively. State and prove an appropriate version of Jensens formula in this case. 3. Let n(r) be as in (4.8.5). Show that
r 0

n(t) dt = t

n( r )

ln
j =1

r . |aj |

Ch: 1

97 (4-36)

TOC

Index

4.9. ANALYTIC CONTINUATION 4. With f as in (4.8.5) and M (r) = max{|f (z )| : |z | = r}, show that for 0 < r < R,
r 0

37

n(t) M (r) . dt ln ( k ) t |f (0)|rk /k !

5. Let f be as in (4.8.5). Show that the function r 1 2

2 0

ln |f (reit )| dt

is increasing, and discuss the nature of its graph on the interval (0, R).

4.9

Analytic Continuation

In this section we examine the problem of extending an analytic function to a larger domain. An example of this has already been encountered in the Schwarz reection principle (2.2.15). We rst consider a function dened by a power series.

4.9.1

Denition

Let f (z ) = n=0 an (z z0 )n have radius of convergence r, 0 < r < . Let z be a point such that |z z0 | = r and let r(t) be the radius of convergence of the expansion of f about the point z1 = (1 t)z0 + tz , 0 < t < 1. Then r(t) (1 t)r (Figure 4.9.1). If

o z*

z z0
o

(1

-t

)r

tr

Figure 4.9.1 r(t) = (1 t)r for some (hence for all) t (0, 1), so that there is no function g analytic on an open set containing D(z0 , r) {z } and such that g = f on D(z0 , r), then z is said to be a singular point of f . Equivalently, z D(0, r) is not a singular point of f i f has an analytic extension to an open set containing D(z0 , r) {z }. We are going to show that there is always at least one singular point on the circle of convergence, although in general, its exact location will not be known. Before doing this, we consider a special case in which it is possible to locate a singular point.

Ch: 1

98 (4-37)

TOC

Index

38

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

4.9.2

Theorem

In (4.9.1), if an is real and nonnegative for all n, then z0 + r is a singular point. Proof. Fix z1 between z0 and z0 + r. Note that since an 0 for all n and z1 z0 is a positive real number, f and its derivatives are nonnegative at z1 . Now assume, to the contrary, that the Taylor series expansion of f about z1 does converge for some z2 to the right of z0 + r. Then we have

k=0

f (k) (z1 ) (z2 z1 )k < +. k!

But by the remark after (2.2.18),

f (k) (z1 ) =
n=k

n(n 1) (n k + 1)an (z1 z0 )nk

for k = 0, 1, 2, . . . . Substituting this for f (k) (z1 ) in the Taylor expansion of f about z1 and using the fact that the order of summation in a double series with nonnegative terms can always be reversed, we get

+ > =

k=0

n=k

n(n 1) (n k + 1)an (z1 z0 )nk n an (z1 z0 )nk (z2 z1 )k k

n

(z2 z1 )k k!

k=0

n=k

=
n=0

an
k=0

n (z1 z0 )nk (z2 z1 )k k

=
n=0

an (z2 z0 )n
n=0

by the binomial theorem. But this implies that greater than r, a contradiction.

an (z z0 )n has radius of convergence

The preceding theorem is illustrated by the geometric series 1 + z + z 2 + , which has radius of convergence equal to 1 and which converges to 1/(1 z ) for |z | < 1. In this case, z = 1 is the only singular point, but as we will see later, the other extreme is also possible, namely that every point on the circle of convergence is a singular point.

4.9.3

Theorem

In (4.9.1), let = {z : |z z0 | = r} be the circle of convergence. Then there is at least one singular point on . Proof. If z is not a singular point, then there is a function fz analytic on a disk D(z, z ) such that fz = f on D(z0 , r) D(z, z ). Say there are no singular points on .

Ch: 1

99 (4-38)

TOC

Index

4.9. ANALYTIC CONTINUATION

39

By compactness, is covered by nitely many such disks, say by D(zj , j ), j = 1, . . . , n. Dene f (z ), z D(z0 , r) fzj (z ), z D(zj , j ), j = 1, . . . , n.

g (z ) =

We show that g is well dened. If D(zj , j ) D(zk , k ) = , then also D(zj , j ) D(zk , k ) D(z0 , r) = , as is veried by drawing a picture. Now fzj fzk = f f = 0 on D(zj , j ) D(zk , k ) by the identity theorem (2.4.8), proving that g is well dened. Thus g is analytic on D(z0 , s) for some s > r, and the Taylor expansion of g about z0 coincides with that of f since g = f on D(z0 , r). This means that the expansion of f converges in a disk of radius greater than r, a contradiction. We are now going to construct examples of power series for which the circle of convergence is a natural boundary, that is, every point on the circle of convergence is a singular point. The following result will be needed.

4.9.4

Lemma

Let f1 (w) = (wp + wp+1 )/2, p a positive integer. Then |w| < 1 implies |f1 (w)| < 1, and if = D(0, 1) D(1, ), > 0, then f1 (D(0, r)) for some r > 1. Proof. If |w| 1, then |f1 (w)| = |w|p |1 + w|/2 |1 + w|/2, which is less than 1 unless w = 1, in which case f1 (w) = 1. Thus |w| < 1 implies |f1 (w)| < 1, and f1 (D(0, 1)) . 1 Hence f1 () is an open set containing D(0, 1). Consequently, there exists r > 1 such 1 1 that D(0, r) f1 (), from which it follows that f1 (D(0, r) f1 (f1 ()) . The construction of natural boundaries is now possible.

4.9.5

Hadamard Gap Theorem

Suppose that f (z ) = k=1 ak z nk and, for some s > 1, nk+1 /nk s for all k . (We say that k ak z nk is a gap series.) If the radius of convergence of the series is 1, then every point on the circle of convergence is a singular point. Proof. We will show that 1 is a singular point, from which it will follow (under these hypotheses) that every point on the unit circle is a singular point. Thus assume, to the contrary, that 1 is not a singular point. Then, for some > 0, f has an analytic extension g to D(0, 1) D(1, ). Let p be a positive integer such that s > (p + 1)/p, and let f1 and r > 1 be as in Lemma 4.9.4. Then h(w) = g (f1 (w)) is analytic on D(0, r), and for

Ch: 1

100 (4-39)

TOC

Index

40 |w| < 1,

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

g (f1 (w)) = f (f1 (w)) =

k=1

ak (f1 (w))nk ak 2nk (wp + wp+1 )nk

k=1 nk

= =
k=1

ak 2nk
n=0 nk

nk wp(nk n) w(p+1)n n nk wpnk +n . n

=
k=1

ak 2nk
n=0

Now for each k we have nk+1 /nk s > (p + 1)/p, so pnk + nk < pnk+1 . Therefore, the n nk pnk +n highest power of w that appears in nk , namely wpnk +nk , is less than the =0 n w n k+1 nk+1 pnk+1 +n pnk+1 w . This means that the series lowest power w that appears in n=0 n
nk

ak 2nk
k=1 n=0

nk wpnk +n n

is (with a grouping of terms) precisely the Taylor expansion of h about w = 0. But since h is analytic on D(0, r), this expansion converges absolutely on D(0, r), hence (as there are no repetition of powers of w),
nk

k=1

|ak |2

nk n=0

nk |w|pnk +n < , n

that is [as in the above computation of g (f1 (w))],

k=1

|ak |2nk (|w|p + |w|p+1 )nk <

nk

sequently, ak z converges for some z with |z | > 1, contradicting the assumption that the series dening f has radius of convergence 1. Finally, if z = ei is not a singular point, let q (z ) = f (ei z ) = k=1 ak eink z nk (with ink radius of convergence 1, as before, because |e | = 1). Now f extends to a function analytic on D(0, 1) D(z , ) for some > 0, and thus q extends to a function analytic on D(0, 1) D(1, ), contradicting the above argument. Some typical examples of gap series are
k=1

|w | ) for |w| < r. But if 1 < |w| < r, then 2nk (|w|p + |w|p+1 )nk = |w|p ( 1+2 k=1 nk

> 1. Con-

z 2 and

k=1

z k! .

Remarks
The series n=0 z n diverges at every point of the circle of convergence since |z |n does not approach 0 when |z | = 1. However, z = 1 is the only singular point since (1 z )1

Ch: 1

101 (4-40)

TOC

Index

4.9. ANALYTIC CONTINUATION is analytic except at z = 1. On the other hand, for if ak = 0, k = 2n ; a2n = 1/n!, then
n

41
1 2n n=1 n! z

has radius of convergence 1,

n

lim sup |an |1/n = lim sup |a2n |1/2 = lim sup(1/n!)1/2 = 1
n n n

because ln[(n!)1/2 ] = 2n ln(n!) = 2n

k=1
n

ln k 2n n ln n 0.

The series converges (as does every series obtained from it by termwise dierentiation) at each point of the circle of convergence, and yet by (4.9.5), each such point is singular. The conclusion of Theorem 4.9.5 holds for any (nite) radius of convergence. For if ak z nk has radius of convergence r, then ak (rz )nk has radius of convergence 1. We now consider chains of functions dened by power series.

4.9.6

Denitions

A function element in is a pair (f, D), where D is a disk contained in and f is analytic on D. (The convention D = D(0, 1) is no longer in eect.) If z is an element of D, then (f, D) is said to be a function element at z . Two function elements (f1 , D1 ) and (f2 , D2 ) in are direct analytic continuations of each other (relative to ) if D1 D2 = and f1 = f2 on D1 D2 . Note that in this case, f1 f2 is an extension of f1 (respectively f2 ) from D1 (respectively D2 ) to D1 D2 . If there is a chain (f1 , D1 ), (f2 , D2 ), . . . , (fn , Dn ) of function elements in , with (fi , Di ) and (fi+1 , Di+1 ) direct analytic continuations of each other for i = 1, 2, . . . , n 1, then (f1 , D1 ) and (fn , Dn ) are said to be analytic continuations of each other relative to . Suppose that is a curve in , with dened on the interval [a, b]. If there is a partition a = t0 < t1 < < tn = b, and a chain(f1 , D1 ), . . . , (fn , Dn ) of function elements in such that (fi+1 , Di+1 ) is a direct analytic continuation of (fi , Di ) for i = 1, 2, . . . , n 1, and (t) Di for ti1 t ti , i = 1, 2, . . . , n, then (fn , Dn ) is said to be an analytic continuation of (f1 , D1 ) along the curve .

4.9.7

Theorem

Analytic continuation of a given function element along a given curve is unique, that is, if (fn , Dn ) and (gm , Em ) are two continuations of (f1 , D1 ) along the same curve , then fn = gm on Dn Em .

Proof. Let the rst continuation be (f1 , D1 ), . . . , (fn , Dn ), and let the second continuation be (g1 , E1 ), . . . , (gm , Em ), where g1 = f1 , E1 = D1 . There are partitions a = t0 < t1 < < tn = b, a = s0 < s1 < < sm = b such that (t) Di for ti1 t ti , i = 1, 2, . . . , n, (t) Ej for sj 1 t sj , j = 1, 2, . . . , m. We claim that if 1 i n, 1 j m, and [ti1 , ti ] [sj 1 , sj ] = , then (fi , Di ) and (gj , Ej ) are direct analytic continuations of each other. This is true when i = j = 1, since g1 = f1 and E1 = D1 . If it is not true for all i and j , pick from all (i, j ) for which the

Ch: 1

102 (4-41)

TOC

Index

42

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

statement is false a pair such that i + j is minimal. Say ti1 sj 1 [then i 2, for if i = 1, then sj 1 = t0 = a, hence j = 1, and we know that the result holds for the pair (1,1)]. We have ti1 sj since [ti1 , ti ] [sj 1 , sj ] = , hence sj 1 ti1 sj . Therefore (ti1 ) Di1 Di Ej , in particular, this intersection is not empty. Now (fi , Di ) is a direct analytic continuation of (fi1 , Di1 ), and furthermore (fi1 , Di1 ) is a direct analytic continuation of (gj , Ej ) by minimality of i + j (note that ti1 [ti2 , ti1 ] [sj 1 , sj ], so the hypothesis of the claim is satised). Since Di1 Di Ej = , (fi , Di ) must be a direct continuation of gj , Ej ), a contradiction. Thus the claim holds for all i and j , in particular for i = n and j = m. The result follows.

4.9.8

Denition

Let be an open connected subset of C. The function elements (f1 , D1 ) and (f2 , D2 ) in are said to be equivalent if they are analytic continuations of each other relative to . (It is immediate that this is an equivalence relation.) An equivalence class of function elements in such that for every z there is an element (f, D) with z D is called a generalized analytic function on . Note that connectedness of is necessary in this denition if there are to be any generalized analytic functions on at all. For if z1 , z2 , there must exist equivalent function elements (f1 , D1 ) and (f2 , D2 ) at z1 and z2 respectively. This implies that there is a curve in joining z1 to z2 . Note also that if g is analytic on , then g determines a generalized analytic function on in the following sense. Take = {(f, D) : D and f = g |D }. However, not every generalized analytic function arises from a single analytic function in this way (see Problem 2). The main result of this section, the monodromy theorem (4.9.11), addresses the question of when a generalized analytic function is determined by a single analytic function.

4.9.9

Denition

Let 0 and 1 be curves in a set S C (for convenience, let 0 and 1 have common domain [a, b]). Assume 0 (a) = 1 (a) = z0 , 0 (b) = 1 (b) = z1 , that is, the curves have the same endpoints. Then 0 and 1 are said to be homotopic (in S ) if there is a continuous map H : [a, b] [0, 1] S (called a homotopy of 0 and 1 ) such that H (t, 0) = 0 (t), H (t, 1) = 1 (t), a t t; H (a, s) = z0 , H (b, s) = z1 , 0 s 1. Intuitively, H deforms 0 into 1 without moving the endpoints or leaving the set S . For 0 s 1, the curve t H (t, s) represents the state of the deformation at time s.

4.9.10

Theorem

Let be an open connected subset of C, and let 0 , 1 be curves in that are homotopic in . Let (f, D) be a function element at z0 , the initial point of 0 and 1 . Assume that

Ch: 1

103 (4-42)

TOC

Index

4.9. ANALYTIC CONTINUATION

43

(f, D) can be continued along all possible curves in , that is, if is a curve in joining z0 to another point zn , there is an analytic continuation (fn , Dn ) of (f, D) along . If (g0 , D0 ) is a continuation of (f, D) along 0 and (g1 , D1 ) is a continuation of (f, D) along 1 , then g0 = g1 on D0 D1 . (Note that D0 D1 = since the terminal point z1 of 0 and 1 belongs to D0 D1 .) Thus (g0 , D0 ) and (g1 , D1 ) are direct analytic continuations of each other. Proof. Let H be a homotopy of 0 and 1 . By hypothesis, if 0 s 1, then (f, D) can be continued along the curve s = H ( , s), say to (gs , Ds ). Fix s and pick one such continuation, say (h1 , E1 ), . . . , (hn , En ) [with (h1 , E1 ) = (f, D), (hn , En ) = (gs , Ds )]. There is a partition a = t0 < t1 < tn = b such that s (t) Ei for ti1 t ti , i = 1, . . . , n. Let Ki be the compact set s ([ti1 , ti ]) Ei , and let = min {dist(Ki , C \ Ei } > 0.
1in

Since H is uniformly continuous, there exists > 0 such that if |s s1 | < , then |s (t) s1 (t)| < for all t [a, b]. In particular, if ti1 t ti , then since s (t) Ki and |s (t) s1 (t)| < , we have s1 (t) Ei . Thus by denition of continuation along a curve, (h1 , E1 ), . . . , (hn , En ) is also a continuation of (f, D) along s1 . But we specied at the beginning of the proof that (f, D) is continued along s1 to (gs1 , Ds1 ). By (4.9.7), gs = gs1 on Ds Ds1 . Thus for each s [0, 1] there is an open interval Is such that gs = gs1 on Ds Ds1 whenever s1 Is . Since [0, 1] can be covered by nitely many such intervals, it follows that g0 = g1 on D0 D1 .

4.9.11

Monodromy Theorem

Let be an open connected subset of C with the property that every closed curve in is homotopic to a point, that is, homotopic (in ) to 0 z , where z is the initial and terminal point of . Let be a generalized analytic function on , and assume that each element of can be continued along all possible curves in . Then there is a function g analytic on such that whenever (f, D) we have g = f on D. Thus is determined by a single analytic function. Proof. If z there is a function element (f, D) such that z D. Dene g (z ) = f (z ). We must show that g is well dened. If (f , D ) and z D , we have to show that f (z ) = f (z ). But since (f, D), (f , D ) , there is a continuation in from (f, D) to (f , D ); since z D D , we can nd a curve (in fact a polygonal path) in with initial and terminal point z such that the continuation is along . But by hypothesis, is homotopic to the curve 0 z . Since (f, D) is a continuation of (f, D) along 0 , it follows from (4.9.10) that f = f on D D , in particular, f (z ) = f (z ). Since g = f on D, g is analytic on .

Remarks
Some authors refer to (4.9.10), rather than (4.9.11), as the monodromy theorem. Still others attach this title to our next result (4.9.13), which is a corollary of (4.9.11). It is

Ch: 1

104 (4-43)

TOC

Index

44

CHAPTER 4. APPLICATIONS OF THE CAUCHY THEORY

appropriate at this point to assign a name to the topological property of that appears in the hypothesis of (4.9.11).

4.9.12

Denition

Let be a plane region, that is, an open connected subset of C. We say that is (homotopically) simply connected if every closed curve in is homotopic (in ) to a point. In the next chapter, we will show that the homotopic and homological versions of simple connectedness are equivalent. Using this terminology, we have the following corollary to the monodromy theorem (4.9.11).

4.9.13

Theorem

Let be simply connected region and let (f, D) be a function element in such that (f, D) can be continued along all curves in whose initial points are in D. Then there is an analytic function g on such that g = f on D. Proof (outline). Let be the collection of all function elements (h, E ) such that (h, E ) is a continuation of (f, D). One can then verify that satises the hypothesis of (4.9.11). Since (f, D) , the result follows.

Alternatively, we need not introduce at all, but instead imitate the proof of (4.9.11).

We conclude this section with an important and interesting application of analytic continuation in simply connected regions.

4.9.14

Theorem

If is a (homotopically) simply connected region, then every harmonic function on has a harmonic conjugate. Proof. If u is harmonic on , we must produce an analytic function g on such that u = Re g . We make use of previous results for disks; if D is a disk contained in , then by (1.6.2), there is an analytic function f on D such that Re f = u. That is, (f, D) is a function element in with Re f = u on D. If : [a, b] is any curve in such that (a) D, we need to show that (f, D) can be continued along . As in the proof of (3.1.7), there is a partition a = t0 < t1 < < tn = b and disks D1 , . . . , Dn with centers at (t1 ), . . . , ( tn ) respectively, such that if tj 1 t tj , then (t) Dj . Now D D1 = , and by repeating the above argument we see that there exists f1 analytic on D1 such that Re f1 = u on D1 . Since f f1 is pure imaginary on D D1 , it follows (from the open mapping theorem (4.3.1), for example) that f f1 is a purely imaginary constant on D D1 . By adding this constant to f1 on D1 , we obtain a new f1 on D1 such that (f1 , D1 ) is a direct continuation of (f, D). Repeating this process with (f1 , D1 ) and (f2 , D2 ), and so on, we obtain a continuation (fn , Dn ) of (f, D) along . Thus by (4.9.13), there is an analytic function g on such that g = f on D. Then Re g = u on D, and hence by the identity theorem for harmonic functions (2.4.14), Re g = u on .

Ch: 1

105 (4-44)

TOC

Index

4.9. ANALYTIC CONTINUATION

45

In the next chapter we will show that the converse of (4.9.14) holds. However, this will require a closer examination of the connection between homology and homotopy. Also, we can give an alternative (but less constructive) proof of (4.9.14) after proving the Riemann mapping theorem.

Problems
n! 1. Let f (z ) = n=0 z , z D (0, 1). Show directly that f has C (0, 1) as its natural boundary without appealing to the Hadamard gap theorem. (Hint: Look at f on radii which terminate at points of the form ei2p/q where p and q are integers.)

2. Let f (z ) = Log z = n=1 (1)n1 (z 1)n /n, z D = D(1, 1). Let = C \ {0} and let be the equivalence class determined by (f, D). (a) Show that is actually a generalized analytic function on , that is, if z then there is an element (g, E ) with z E . (b) Show that there is no function h analytic on such that for every (g, E ) we have h = g on E . 3. Criticize the following argument: Let f (z ) = n=0 an (z z0 )n have radius of convergence r. If z1 D(z0 , r), then by the rearrangement procedure of (4.9.2) we can nd the Taylor expansion of f about z1 , namely

f (z ) =
k=0 n=k

n an (z1 z0 )nk (z z1 )k . k

If the expansion about z1 converges at some point z / D(z0 , r), then since power series converge absolutely inside the circle of convergence, we may rearrange the expansion about z1 to show that the original expansion about z0 converges at z , a contradiction. Consequently, for any function dened by a power series, the circle of convergence is a natural boundary. 4. (Law of permanence of functional equations). Let F : Ck+1 C be such that F and all its rst order partial derivatives are continuous. Let f1 , . . . , fk be analytic on a disk D, and assume that F (z, f1 (z ), . . . , fk (z )) = 0 for all z D. Let (fi1 , D1 ), (fi2 , D2 ), . . . , (fin , Dn ), with fi1 = fi , D1 = D, form a continuation of (fi , D), i = 1, . . . , k. Show that F (z, f1n (z ), . . . , fkn (z )) = 0 for all z Dn . An example: If eg = f on D and the continuation carries f into f and g into g , then eg = f on Dn (take F (z, w1 , w2 ) = w1 ew2 , f1 = f, f2 = g ).

5. Let (f , D ) be a continuation of (f, D). Show that (f , D ) is a continuation of (f , D). (The derivative of the continuation, that is, f , is the continuation of the derivative.)

Reference
W. Rudin, Real and Complex Analysis, 3rd ed., McGraw Hill Series in Higher Mathematics, New York, 1987.

Ch: 1

106 (4-45)

TOC

Index

Chapter 5

Families of Analytic Functions

In this chapter we consider the linear space A() of all analytic functions on an open set and introduce a metric d on A() with the property that convergence in the d-metric is uniform convergence on compact subsets of . We will characterize the compact subsets of the metric space (A(), d) and prove several useful results on convergence of sequences of analytic functions. After these preliminaries we will present a fairly standard proof of the Riemann mapping theorem and then consider the problem of extending the mapping function to the boundary. Also included in this chapter are Runges theorem on rational approximations and the homotopic version of Cauchys theorem.

5.1
5.1.1

The Spaces A() and C ()

Denitions

Let be an open subset of C. Then A() will denote the space of analytic functions on , while C() will denote the space of all continuous functions on . For n = 1, 2, 3 . . . , let Kn = D(0, n) {z : |z w| 1/n for all w C \ }. By basic topology of the plane, the sequence {Kn } has the following three properties: (1) Kn is compact, o (2) Kn Kn +1 (the interior of Kn+1 ), (3) If K is compact, then K Kn for n suciently large. Now x a nonempty open set , let {Kn } be as above, and for f, g C(), dene

d(f, g ) =
n=1

1 2n

f g Kn , 1 + f g Kn Kn = Kn =

where f g
Kn

sup{|f (z ) g (z )| : z Kn }, 0, 1

Ch: 1

107 (5-1)

TOC

Index

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.1.2

Theorem

The assignment (f, g ) d(f, g ) denes a metric on C (). A sequence {fj } in C () is dconvergent (respectively d-Cauchy) i {fj } is uniformly convergent (respectively uniformly Cauchy) on compact subsets of . Thus (C (), d) and (A(), d) are complete metric spaces. Proof. That d is a metric on C () is relatively straightforward. The only troublesome part of the argument is verication of the triangle inequality, whose proof uses the inequality: If a, b and c are nonnegative numbers and a b + c, then b c a + . 1+a 1+b 1+c To see this, note that h(x) = x/(1 + x) increases with x 0, and consequently h(a) c b c h(b + c) = 1+b b+c + 1+b+c 1+b + 1+c . Now let us show that a sequence {fj } is d-Cauchy i {fj } is uniformly Cauchy on compact subsets of . Suppose rst that {fj } is d-Cauchy, and let K be any compact subset of . By the above property (3) of the sequence {Kn }, we can choose n so large that K Kn . Since d(fj , fk ) 0 as j, k , the same is true of fj fk Kn . But fj fk K fj fk Kn , hence {fj } is uniformly Cauchy on K . Conversely, assume that {fj } is uniformly Cauchy on compact subsets of . Let > 0 and choose a positive integer m such that n=m+1 2n < . Since {fj } is uniformly Cauchy on Km in particular, there exists N = N (m) such that j, k N implies fj fk Km < , hence
m n=1

1 2n

fj fk Kn 1 + fj fk Kn n=1

1 2n

fj fk
m Km

Kn

fj fk It follows that for j, k N ,

1 < . n 2 n=1

d(fj , fk ) =
n=1

1 2n

fj fk Kn < 2. 1 + fj fk Kn

If {fn } is a sequence in A() and fn f uniformly on compact subsets of , then we know that f A() also. The next few theorems assert that certain other properties of the limit function f may be inferred from the possession of these properties by the fn . The rst results of this type relate the zeros of f to those of the fn .

The remaining statements in (5.1.2) follow from the above, Theorem 2.2.17, and completeness of C.

5.1.3

Hurwitzs Theorem

Suppose that {fn } is a sequence in A() that converges to f uniformly on compact subsets of . Let D(z0 , r) and assume that f (z ) = 0 for |z z0 | = r. Then there is a positive integer N such that for n N , fn and f have the same number of zeros in D(z0 , r).

Ch: 1

108 (5-2)

TOC

Index

5.1. THE SPACES A() AND C ()

Proof. Let = min{|f (z )| : |z z0 | = r} > 0. Then for suciently large n, |fn (z ) f (z )| < |f (z )| for |z z0 | = r. By Rouch es theorem (4.2.8), fn and f have the same number of zeros in D(z0 , r).

5.1.4

Theorem

Let {fn } be a sequence in A() such that fn f uniformly on compact subsets of . If is connected and fn has no zeros in for innitely many n, then either f has no zeros in or f is identically zero. Proof. Assume f is not identically zero, but f has a zero at z0 . Then by the identity theorem (2.4.8), there is r > 0 such that the hypothesis of (5.1.3) is satised. Thus for suciently large n, fn has a zero in D(z0 , r).

5.1.5

Theorem

Let {fn } be a sequence in A() such that fn converges to f uniformly on compact subsets of . If is connected and the fn are one-to-one on , then either f is constant on or f is one-to-one. Proof. Assume that f is not constant on , and choose any z0 . The sequence {fn fn (z0 )} satises the hypothesis of (5.1.4) on the open connected set \{z0 } (because the fn are one-to-one). Since f f (z0 ) is not identically zero on \ {z0 }, it follows from (5.1.4) that f f (z0 ) has no zeros in \ {z0 }. Since z0 is an arbitrary point of , we conclude that f is one-to-one on . The next task will be to identify the compact subsets of the space A() (equipped with the topology of uniform convergence on compact subsets of ). After introducing the appropriate notion of boundedness for subsets F A(), we show that each sequence of functions in F has a subsequence that converges uniformly on compact subsets of . This leads to the result that a subset of A() is compact i it is closed and bounded.

5.1.6

Denition

A set F C () is bounded if for each compact set K , sup{ f K : f F} < , that is, the functions in F are uniformly bounded on each compact subset of . We will also require the notion of equicontinuity for a family of functions.

5.1.7

Denition

A family F of functions on is equicontinuous at z0 if given > 0 there exists > 0 such that if z and |z z0 | < , then |f (z ) f (z0 )| < for all f F . We have the following relationship between bounded and equicontinuous subsets of A().

Ch: 1

109 (5-3)

TOC

Index

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.1.8

Theorem

Let F be a bounded subset of A(). Then F is equicontinuous at each point of . Proof. Let z0 and choose r > 0 such that D(z0 , r) . Then for z D(z0 , r) and f F , we have f (z ) f (z0 ) = Thus |f (z ) f (z0 )| 1 sup 2 1 2i
C (z0 ,r )

1 f (w) dw wz 2i

C (z0 ,r )

f (w) dw. w z0

f (w) f (w) : w C (z0 , r) 2r wz w z0 f (w) : w C (z0 , r) . (w z )(w z0 )

= r|z z0 | sup

But by hypothesis, there exists Mr such that |f (w)| Mr for all w C (z0 , r) and all f F . Consequently, if z D(z0 , r/2) and f F , then r|z z0 | sup f (w) : w C (z0 , r) (w z )(w z0 ) r|z z0 | Mr , (r/2)2

proving equicontinuity of F . We will also need the following general fact about equicontinuous families.

5.1.9

Theorem

Suppose F is an equicontinuous subset of C () (that is, each f F is continuous on and F is equicontinuous at each point of ) and {fn } is a sequence from F such that fn converges pointwise to f on . Then f is continuous on and fn f uniformly on compact subsets of . More generally, if fn f pointwise on a dense subset of , then fn f on all of and the same conclusion holds. Proof. Let > 0. For each w , choose a w > 0 such that |fn (z ) fn (w)| < for each z D(w, w ) and all n. It follows that |f (z ) f (w)| for all z D(w, w ), so f is continuous. Let K be any compact subset of . Since {D(w, w ) : w K } is an open cover of K , there are w1 , . . . , wm K such that K m j =1 D (wj , wj ). Now choose N such that n N implies that |f (wj ) fn (wj )| < for j = 1, . . . , m. Hence if z D(wj , wj ) and n N , then In particular, if z K and n N , then |f (z ) fn (z )| < 3, showing that fn f uniformly on K . Finally, suppose only that fn f pointwise on a dense subset S . Then as before, |fn (z ) fn (w)| < for all n and all z D(w, w ). But since S is dense, D(w, w ) contains a point z S , and for m and n suciently large, Thus {fn (w)} is a Cauchy sequence and therefore converges, hence {fn } converges pointwise on all of and the rst part of the theorem applies. |fm (w) fn (w)| |fm (w) fm (z )| + |fm (z ) fn (z )| + |fn (z ) fn (w)| < 3. |f (z ) fn (z )| |f (z ) f (wj )| + |f (wj ) fn (wj )| + |fn (wj ) fn (z )| < 3.

Ch: 1

110 (5-4)

TOC

Index

5.1. THE SPACES A() AND C ()

5.1.10

Montels Theorem

Let F be a bounded subset of A(), as in (5.1.6). Then each sequence {fn } from F has a subsequence {fnj } which converges uniformly om compact subsets of .

Remark
A set F C ( is said to be relatively compact if the closure of F in C () is compact. The conclusion of (5.1.10) is equivalent to the statement that F is a relatively compact subset of C (), and hence of A(). Proof. Let {fn } be any sequence from F and choose any countable dense subset S = {z1 , z2 , . . . } of . The strategy will be to show that fn } has a subsequence which converges pointwise on S . Since F is a bounded subset of A(), it is equicontinuous on by (5.1.8). Theorem 5.1.9 will then imply that this subsequence converges uniformly on compact subsets of , thus completing the proof. So consider the following bounded sequences of complex numbers:
{fj (z1 )} j =1 , {fj (z2 )}j =1 , . . . . There is a subsequence {f1j } j =1 of {fj }j =1 which converges at z1 . There is a subsequence {f2j }j =1 of {f1j }j =1 which converges at z2 and (necessarily) at z1 as well. Proceeding inductively, for each n 1 and each k = 1, . . . , n we construct sequences {fkj } j =1 converging at z1 , . . . , zk , each a subsequence of the preceding sequence.

Put gj = fjj . Then {gj } is a subsequence of {fj }, and {gj } converges pointwise on {z1 , z2 , . . . } since for each n, {gj } is eventually a subsequence of {fnj } j =1 .

5.1.11

Theorem (Compactness Criterion)

Let F A(). Then F is compact i F is closed and bounded. Also, F is relatively compact i F is bounded. (See Problem 3 for the second part of this theorem.) Proof. If F is compact, then F is closed (a general property that holds in any metric space). In order to show that F is bounded, we will use the following device. Let K be any compact subset of . Then f f K is a continuous map from A() into R. Hence { f K : f F} is a compact subset of R and thus is bounded. Conversely, if F is closed and bounded, then F is closed and, by Montels theorem, relatively compact. Therefore F is compact.

Remark
Problem 6 gives an example which shows that the preceding compactness criterion fails in the larger space C (). That is, there are closed and bounded subsets of C () that are not compact.

Ch: 1

111 (5-5)

TOC

Index

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.1.12

Theorem

Suppose F is a nonempty compact subset of A(). Then given z0 , there exists g F such that |g (z0 )| |f (z0 )| for all f F . Here is a compactness result that will be needed for the proof of the Riemann mapping theorem in the next section. Proof. Just note that the map f |f (z0 )|, f A(), is continuous.

5.1.13

Theorem

Assume that is connected, z0 , and > 0. Dene F = {f A() : f is a one-to-one map of into D(0, 1) and |f (z0 )| }. Proof. By its denition, F is bounded, and F is closed by (5.1.5). Thus by (5.1.11), F is compact. To prove the last statement of the theorem, note that if fn F and fn f uniformly on compact subsets of , then (5.1.5) would imply that f F , were it not for the annoying possibility that |f (w)| = 1 for some w . But if this happens, the maximum principle implies that f is constant, contradicting |f (z0 )| > 0. Then F is compact. The same conclusion holds with D(0, 1) replaced by D(0, 1).

The nal result of this section shows that if is connected, then any bounded sequence in A() that converges pointwise on a set having a limit point in , must in fact converge uniformly on compact subsets of .

5.1.14

Vitalis Theorem

Let {fn } be a bounded sequence in A() where is connected. Suppose that {fn } converges pointwise on S and S has a limit point in . Then {fn } is uniformly Cauchy on compact subsets of , hence uniformly convergent on compact subsets of to some f A().

Proof. Suppose, to the contrary, that there is a compact set K such that {fn } is not uniformly Cauchy on K . Then for some > 0, we can nd sequences {mj } and {nj } of positive integers such that m1 < n1 < m2 < n2 < and for each j , fmj fnj K . Put {gj } = {fmj } and {hj } = {fnj }. Now apply Montels theorem (5.1.10) to {gj } to obtain a subsequence {gjr } converging uniformly on compact subsets of to some g A(), and then apply Montels theorem to {hjr } to obtain a subsequence converging uniformly on compact subsets of to some h A(). To prevent the notation from getting out of hand, we can say that without loss of generality, we have gn g and hn h uniformly on compact subsets, and gn hn K for all n, hence g h K . But by hypothesis, g = h on S and therefore, by (2.4.9), g = h on , a contradiction.

Problems
1. Let F = {f A(D(0, 1)) : Re f > 0 and |f (0)| 1}. Prove that F is relatively compact. Is F compact? (See Section 4.6, Problem 2.)

Ch: 1

112 (5-6)

TOC

Index

5.1. THE SPACES A() AND C ()

2. Let be (open and) connected and let F = {f A() : |f (z ) a| r for all z }, where r > 0 and a C are xed. Show that F is a normal family, that is, if fn F , n = 1, 2, . . . , then either there is a subsequence {fnj } converging uniformly on compact subsets to a function f A() or there is a subsequence {fnj } converging uniformly on compact subsets to . (Hint: Look at the sequence {1/(fn a)}.) 3. (a) If F C (), show that F is relatively compact i each sequence in F has a convergent subsequence (whose limit need not be in F ). (b) Prove the last statement in Theorem 5.1.11. 4. Let F A(D(0, 1)). Show that F is relatively compact i there is a sequence of nonnegative real numbers Mn with lim supn (Mn )1/n 1 such that for all f F and all n = 0, 1, 2, . . . , we have |f (n) (0)/n!| Mn . |f (a)|2 1 R2
2 0 0 R

5. (a) Suppose that f is analytic on and D(a, R) . Prove that |f (a + reit )|2 r dr dt.

(b) Let M > 0 and dene F to be the set {f A() : Show that F is relatively compact.

|f (x + iy )|2 dx dy M }.

6. Let be open and K = D(a, R) . Dene F to be the set of all f C () such that |f (z )| 1 for all z and f (z ) = 0 for z \ K }. Show that F is a closed and bounded subset of C (), but F is not compact. (Hint: Consider the map from F to the reals given by
1

(1 |f (x + iy )|) dx dy

Show that this map is continuous but not bounded on F .)

.ei
0

S (,)

Figure 5.1.1

Ch: 1

113 (5-7)

TOC

Index

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

7. (An application of Vitalis theorem.) Let f be a bounded analytic function on D(0, 1) with the property that for some , f (rei ) approaches a limit L as r 1 . Fix (0, /2) and consider the region S (, ) in Figure 5.1.1. Prove that if z S (, ) and z ei , then f (z ) L. (Suggestion: Look at the sequence of functions dened 1 by fn (z ) = f (ei + n (z ei )), z D(0, 1).) 8. Let L be a multiplicative linear functional on A(), that is, L : A() C such that L(af + bg ) = aL(f ) + bL(g ) and L(f g ) = L(f )L(g ) for all a, b C, f, g A(). Assume L 0. Show that L is a point evaluation, that is, there is some z0 such that L(f ) = f (z0 ) for all f A(). Outline: First show that for f 1, L(f ) = 1. Then apply L to the function I (z ) = z , the identity on , and show that if L(I ) = z0 , then z0 . Finally, if f A(), apply L to the function g (z ) =
f (z )f (z0 ) , z z0 f (z0 ),

z = z0 z = z0 .

9. (Osgoods theorem). Let {fn } be a sequence in A() such that fn f pointwise on . Show that there is an open set U , dense in , such that fn f uniformly on compact subsets of U . In particular, f is analytic on a dense open subset of . (Let An = {z : |fk (z )| n for all k = 1, 2, . . . }. Recall the Baire category theorem: If a complete metric space X is the union of a sequence {Sn } of closed subsets, then some Sn contains a nonempty open ball. Use this result to show that some An contains a disk D. By Vitalis theorem, fn f uniformly on compact subsets of D. Take U to be the union of all disks D such that fn f uniformly on compact subsets of D.)

5.2

Riemann Mapping Theorem

Throughout this section, will be a nonempty open connected proper subset of C with the property that every zero-free analytic function has an analytic square root. Later in the section we will prove that any open subset such that every zero-free analytic function on has an analytic square root must be (homotopically) simply connected, and conversely. Thus we are considering open, connected and simply connected proper subsets of C. Our objective is to prove the Riemann mapping theorem, which states that there is a one-to-one analytic map of onto the open unit disk D. The proof given is due to Fejer and F.Riesz.

5.2.1

Lemma

There is a one-to-one analytic map of into D. Proof. Fix a C \ . Then the function z a satises our hypothesis on and hence there exists h A() such that (h(z ))2 = z a, z . Note that h is one-to-one and 0 / h(). Furthermore, h() is open by (4.3.1), the open mapping theorem, hence so is h() = {h(z ) : z }, and [h(] [h(] = (because 0 / h()). Now choose w h(). Since h() is open, there exists r > 0 such that D(w, r) h(), hence h() D(w, r) = . The function f (z ) = 1/(h(z ) w), z , is one-to-one, and its magnitude is less than 1/r on . Thus rf is a one-to-one map of into D.

Ch: 1

114 (5-8)

TOC

Index

5.2. RIEMANN MAPPING THEOREM

5.2.2

Riemann Mapping Theorem

Let be as in (5.2.1), that is, a nonempty, proper, open and connected subset of C such that every zero-free analytic function on has an analytic square root. Then there is a one-to-one analytic map of onto D. Proof. Fix z0 and a one-to-one analytic map f0 of into D [f0 exists by (5.2.1)]. Let F be the set of all f A() such that f is a one-to-one analytic map of into D and |f (z0 )| |f0 (z0 )|. Note that |f0 (z0 )| > 0 by (4.3.1).

Then F = (since f0 F ) and F is bounded. Also, F is closed, for if {fn } is a sequence in F such that fn f uniformly on compact subsets of , then by (5.1.5), either f is constant on or f is one-to-one. But since fn f , it follows that |f (z0 )| |f0 (z0 )| > 0, so f is one-to-one. Also, f maps into D (by the maximum principle), so f F . Since F is closed and bounded, it is compact (Theorem 5.1.1). Hence by (5.1.2), there exists g F such that |g (z0 )| |f (z0 )| for all f F . We will now show that such a g must map onto D. For suppose that there is some a D \ g (). Let a be as in (4.6.1), that is, a (z ) = za , z D. 1 az

Then a g : D and a g is one-to-one with no zeros in . By hypothesis, there is an analytic square root h for a g . Note also that h2 = a g is one-to-one, and therefore so is h. Set b = h(z0 ) and dene f = b h. Then f (z0 ) = b (b) = 0 and we can write
2 g = a h2 = a (b f )2 = a (2 b f ) = (a b ) f.

Now
g (z0 ) = (a 2 b ) (f (z0 ))f (z0 )

(1)

= (a

2 b ) (0)f (z0 ).

The function a 2 b is an analytic map of D into D , but it is not one-to-one; indeed, it is two-to-one. Hence by the Schwarz-Pick theorem (4.6.3), part (ii), it must be the case that
2 2 |(a 2 b ) (0)| < 1 |a b (0)| .

Since f (z0 ) = 0, it follows from (1) that

2 |g (z0 )| < (1 |a 2 b (0)| )|f (z0 )| |f (z0 )|.

This contradicts our choice of g F as maximizing the numbers |f (z0 )|, f F . Thus g () = D as desired.

Ch: 1

115 (5-9)

TOC

Index

10

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.2.3

Remarks

(a) Any function g that maximizes the numbers {|f (z0 )| : f F} must send z0 to 0. Proof. Let a = g (z0 ). Then a g is a one-to-one analytic map of into D. Moreover,
|(a g ) (z0 )| = | a (g (z0 ))g (z0 )| = | a (a)g (z0 )| 1 = |g (z0 )| 1 |a|2 |g (z0 )| |f0 (z0 )|.

Thus a g F , and since |g (z0 )| maximizes |f (z0 )| for f F , it follows that equality must hold in the rst inequality. Therefore 1/(1 |a|2 ) = 1, so 0 = a = g (z0 ). (b) Let f and h be one-to-one analytic maps of onto D such that f (z0 ) = h(z0 ) = 0 and f (z0 ) = h (z0 ) (it is enough that Arg f (z0 ) = Arg h (z0 )). Then f = h. Proof. The function h f 1 is a one-to-one analytic map of D onto D, and h f 1 (0) = h(z0 ) = 0. Hence by Theorem 4.6.4 (with a = 0), there is a unimodular complex number such that h(f 1 (z )) = z, z D. Thus h(w) = f (w), w D. But if h (z0 ) = f (z0 ) (which is equivalent to Arg h (z0 ) = Arg f (z0 ) since |h (z0 )| = |||f (z0 )| = |f (z0 )|), we have = 1 and f = h. (c) Let f be any analytic map of into D (not necessarily one-to-one or onto) with f (z0 ) = 0. Then with g as in the theorem, |f (z0 )| |g (z0 )|. Also, equality holds i f = g with || = 1. Proof. The function f g 1 is an analytic map of D into D such that f g 1 (0) = 0. By Schwarzs lemma (2.4.16), |f (g 1 (z ))| |z | and |f (g 1 (0)) g (1 z0 ) | 1. Thus |f (z0 )| |g (z0 )|. Also by (2.4.16), equality holds i for some unimodular we have f g 1 (z ) = z , that is, f (z ) = g (z ), for all z D. If we combine (a), (b) and (c), and observe that g (z0 ) will be real and greater than 0 for appropriately chosen unimodular , then we obtain the following existence and uniqueness result. (d) Given z0 , there is a unique one-to-one analytic map g of onto D such that g (z0 ) = 0 and g (z0 ) is real and positive. As a corollary of (d), we obtain the following result, whose proof will be left as an exercise; see Problem 1. (e) Let 1 and 2 be regions that satisfy the hypothesis of the Riemann mapping theorem. Let z1 1 and z2 2 . Then there is a unique one-to-one analytic map f of 1 onto 2 such that f (z1 ) = z2 and f (z1 ) is real and positive. Recall from (3.4.6) that if C and satises any one of the six equivalent conditions listed there, then is called (homologically) simply connected. Condition (6) is that every zero-free analytic function on have an analytic n-th root for n = 1, 2, . . . . Thus if is homologically simply connected, then in particular, assuming = C, the Riemann mapping theorem implies that is conformally equivalent to D, in other words, there is a one-to-one analytic map of onto D. The converse is also true, but before showing this, we need to take a closer look at the relationship between homological simple connectedness and homotopic simple connectedness [see (4.9.12)].

Ch: 1

116 (5-10)

TOC

Index

5.2. RIEMANN MAPPING THEOREM

11

5.2.4

Theorem

Let 0 and 1 be closed curves in an open set C. If 0 and 1 are -homotopic (in other words, homotopic in ), then they are -homologous, that is, n(0 , z ) = n(1 , z ) for every z C \ . Proof. We must show that n(0 , z ) = n(1 , z ) for each z C \ . Thus let z C \ , let H be a homotopy of 0 to 1 , and let be a continuous argument of H z . (See Problem 6 of Section 3.2.) That is, is a real continuous function on [a, b] [0, 1] such that H (t, s) z = |H (t, s) z |ei(t,s) for (t, s) [a, b] [0, 1]. Then for each s [0, 1], the function t (t, s) is a continuous argument of H (, s) z and hence n(H (, s), z ) = (b, s) (a, s) . 2

This shows that the function s n(H (, s), z ) is continuous, and since it is integer valued, it must be constant. In particular, n(H (, 0), z ) = n(H (, 1), z ). In other words, n(0 , z ) = n(1 , z ). The above theorem implies that if is -homotopic to a point in , then must be -homologous to 0. Thus if is a closed path in such that is -homotopic to a point, then f (z ) dz = 0 for every analytic function f on . We will state this result formally.

5.2.5

The Homotopic Version of Cauchys Theorem

Let be a closed path in such that is -homotopic to a point. Then for every analytic function f on .

f (z ) dz = 0

.
a

.
b

Figure 5.2.1

Ch: 1

117 (5-11)

TOC

Index

12

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

Remark
The converse of Theorem 5.2.5 is not true. In particular, there are closed curves and open sets such that is -homologous to 0 but is not homotopic to a point. Take = C \ {a, b}, a = b, and consider the closed path of Figure 5.2.1. Then n(, a) = n(, b) = 0, hence is -homologous to 0. But (intuitively at least) we see that cannot be shrunk to a point without passing through a or b. It follows from this example and Theorem 5.2.4 that the homology version of Cauchys theorem (3.3.1) is actually stronger than the homotopy version (5.2.5). That is, if is a closed path to which the homotopy version applies, then so does the homology version, while the homology version applies to the above path, but the homotopy version does not. However, if every closed path in is homologous to zero, then every closed path is homotopic to a point, as we now show.

5.2.6

Theorem

Let be an open connected subset of C. The following are equivalent. (2) If = C, then is conformally equivalent to D. (3) is homeomorphic to D. (4) is homotopically simply connected. (5) Each closed path in is homotopic to a point. (6) is homologically simply connected. Proof. (1) implies (2): This is the Riemann mapping theorem. (2) implies (3): If = C, this follows because a conformal equivalence is a homeomorphism., while if = C, then the map h(z ) = z/(1 + |z |) is a homeomorphism of C onto D (see Problem 2). (3) implies (4): Let : [a, b] be any closed curve in . By hypothesis there is a homeomorphism f of onto D. Then f is a closed curve in D, and there is a homotopy H (in D) of f to the point f ( (a)) (see Problem 4). Therefore f 1 H is a homotopy in of to (a). (4) implies (5): Every closed path is a closed curve. (5) implies (6): Let be any closed path in . If is -homotopic to a point, then by Theorem 5.2.4, is -homologous to zero. (6) implies (1): This follows from part (6) of (3.4.6). (1) Every zero-free f A() has an analytic square root.

Remark
If is any open set (not necessarily connected) then the statement of the preceding theorem applies to each component of . Therefore (1), (4), (5) and (6) are equivalent for arbitrary open sets. Here is yet another condition equivalent to simple connectedness of an open set .

Ch: 1

118 (5-12)

TOC

Index

5.2. RIEMANN MAPPING THEOREM

13

5.2.7

Theorem

Let be a simply connected open set. Then every harmonic function on has a harmonic conjugate. Conversely, if is an open set such that every harmonic function on has a harmonic conjugate, then is simply connected. Proof. The rst assertion was proved as Theorem 4.9.14 using the method of analytic continuation. However, we can also give a short proof using the Riemann mapping theorem, as follows. First note that we can assume that is connected by applying this case to components. If = C then every harmonic function on has a harmonic conjugate as in Theorem 1.6.2. Suppose then that = C. By the Riemann mapping theorem, there is a conformal equivalence f of onto D. Let u be harmonic on . Then u f 1 is harmonic on D and thus by (1.6.2), there is a harmonic function V on D such that u f 1 + iV is analytic on D. Since (u f 1 + iV ) f is analytic on , there is a harmonic conjugate of u on , namely v = V f . Conversely, suppose that is not simply connected. Then is not homologically simply connected, so there exists z0 C\ and a closed path in such that n(, z0 ) = 0. Thus by (3.1.9) and (3.2.3), the function z z z0 does not have an analytic logarithm on , hence z ln |z z0 | does not have a harmonic conjugate. The nal result of this section is Runges theorem on rational and polynomial approximation of analytic functions. One consequence of the development is another condition that is equivalent to simple connectedness.

5.2.8

Runges Theorem

\ K that contains at least one point Let K be a compact subset of C, and S a subset of C \ K . Dene B (S ) = {f : f is a uniform limit on K of rational in each component of C functions whose poles lie in S }. Then every function f that is analytic on a neighborhood of K is in B (S ). That is, there is a sequence {Rn } of rational functions whose poles lie in S such that Rn f uniformly on K . \ K is Before giving the proof, let us note the conclusion in the special case where C connected. In this case, we can take S = {}, and our sequence of rational functions will actually be a sequence of polynomials. The proof given is due to Sandy Grabiner (Amer. Math. Monthly, 83 (1976), 807-808) and is based on three lemmas.

5.2.9

Lemma

Suppose K is a compact subset of the open set C. If f A(), then f is a uniform limit on K of rational functions whose poles (in the extended plane!) lie in \ K .

5.2.10

Lemma

Let U and V be open subsets of C with V U and V U = . If H is any component of U and V H = , then H V .

Ch: 1

119 (5-13)

TOC

Index

14

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.2.11

Lemma

If K is a compact subset of C and C \ K , then (z )1 B (S ).

Let us see how Runges theorem follows from these three lemmas, and then we will prove the lemmas. First note that if f and g belong to B (S ), then so do f + g and f g . Thus by Lemma 5.2.11 (see the partial fraction decomposition of Problem 4.1.7), \ K belongs to B (S ). Runges theorem is then a every rational function with poles in C consequence of Lemma 5.2.9. (The second of the three lemmas is used to prove the third.)

Proof of Lemma 5.2.9

Let be an open set containing K . By (3.4.7), there is a cycle in \ K such that for every f A() and z K , f (z ) = 1 2i

f (w) dw. wz

Let > 0 be given. Then = dist( , K ) > 0 because and K are disjoint compact sets. Assume [0, 1] is the domain of and let s, t [0, 1], z K . Then f ( (s)) f ( (t)) (t) z (s) z f ( (t))( (s) z ) f ( (s))( (t) z ) = ( (t) z )( (s) z ) f ( (t))( (s) (t)) + (t)(f ( (t)) f ( (s))) z (f ( (t)) f ( (s))) = ( (t) z )( (s) z ) 1 2 (|f ( (t))|| (s) (t)| + | (t)||f ( (t)) f ( (s))| + |z ||f ( (t)) f ( (s))|). Since and f are bounded functions and K is a compact set, there exists C > 0 such that for s, t [0, 1] and z K , the preceding expression is bounded by C (| (s) (t)| + |f ( (t)) f ( (s))|. 2 Thus by uniform continuity of and f on the interval [0, 1], there is a partition 0 = t0 < t1 < < tn = 1 such that for t [tj 1 , tj ] and z K , f ( (t)) f ( (tj )) < . (t) z (tj ) z Dene
n

R(z ) =
j =1

f ( (tj )) ( (tj ) (tj 1 )), z = (tj ). (tj ) z

Ch: 1

120 (5-14)

TOC

Index

5.2. RIEMANN MAPPING THEOREM

15

Then R(z ) is a rational function whose poles are included in the set { (t1 ), . . . , (tn )}, in particular, the poles are in \ K . Now for all z K , |2if (z ) R(z )| = =
j =1 1 tj 1 n

f ( (tj )) f (w) dw ( (tj ) (tj 1 )) wz (tj ) z j =1

tj

f ( (t)) f ( (tj )) (t) z (tj ) z

(t) dt

| (t)| dt = length of .

Since the length of is independent of , the lemma is proved.

Proof of Lemma 5.2.10

Let H be any component of U such that V H = . We must show that H V . let s V H and let G be that component of V that contains s. It suces to show that G = H . Now G H since G is a connected subset of U containing s and H is the union of all subsets with this property. Write H = G (H \ G) = G [(G H ) (H \ G)]. But G H = , because otherwise the hypothesis V U = would be violated. Thus H = G (H \ G), the union of two disjoint open sets. Since H is connected and G = , we have G = H as required.

Proof of Lemma 5.2.11

Suppose rst that S . Then for suciently large |0 |, with 0 in the unbounded component of C \ K , the Taylor series for (z 0 )1 converges uniformly on K . Thus (z 0 )1 B (S ), and it follows that B ((S \ {}) {0 }) B (S ). (If f B ((S \ {}) {0 }) and R is a rational function with poles in (S \ {}) {0 } that approximates f , write R = R1 + R2 where all the poles (if any) of R1 lie in S \ {} and the pole (if any) of R0 is at 0 . But R0 can be approximated by a polynomial P0 , hence R1 + P0 approximates f and has its poles in S , so f B (S ).) Thus it is sucient to establish the lemma for sets S C. We are going to apply Lemma 5.2.10. Put U = C \ K and dene V = { U : (z )1 B (S )}. Recall that by hypothesis, S U and hence S V U . To apply (5.2.10) we must rst show that V is open. Suppose V and is such that 0 < | | < dist(, K ). Then C \ K and for all z K , 1 1 = z (z )[1
z ]

Ch: 1

121 (5-15)

TOC

Index

16

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

Since (z )1 B (S ), it follows from the remarks preceding the proof of Lemma 5.2.9 that (z )1 B (S ). Thus V , proving that V is open. Next well show that V U = . Let w V and let {n } be a sequence in V such that n w. Then as we noted earlier in this proof, |n w| < dist(n , K ) implies w V , so it must be the case that |n w| dist(n , K ) for all n. Since |n w| 0, the distance from w to K must be 0, so w K . Thus w / U , proving that V U = , as desired. Consequently, V and U satisfy the hypotheses of (5.2.10). Let H be any component of U . By denition of S , there exists s S such that s H . Now s V because S V . Thus H V = , and Lemma 5.2.10 implies that H V . We have shown that every component of U is a subset of V , and consequently U V . Since V U , we conclude that U = V .

5.2.12

Remarks

Theorem 5.2.8 is often referred to as Runges theorem for compact sets. Other versions of Runges theorem appear as Problems 6(a) and 6(b). We conclude this section by collecting a long list of conditions, all equivalent to simple connectedness.

5.2.13

Theorem

(b) n(, z ) = 0 for each closed path (or cycle) in and each point z C \ . (c)

If is an open subset of C, the following are equivalent. \ is connected. (a) C

f (z ) dz = 0 for each f A() and each closed path in .

(d) n(, z ) = 0 for each closed curve in and each z C \ . (e) Every analytic function on has a primitive. (f) Every zero-free analytic function on has an analytic logarithm. (g) Every zero-free analytic function on has an analytic n-th root for n = 1, 2, 3, . . . . (h) Every zero-free analytic function on has an analytic square root. (i) is homotopically simply connected. (j) Each closed path in is homotopic to a point. (k) If is connected and = C, then is conformally equivalent to D. (l) If is connected, then is homeomorphic to D. (m) Every harmonic function on has a harmonic conjugate. (n) Every analytic function on can be uniformly approximated on compact sets by polynomials. Proof. See (3.4.6), (5.2.4), (5.2.6), (5.2.7), and Problem 6(b) in this section.

Ch: 1

122 (5-16)

TOC

Index

5.2. RIEMANN MAPPING THEOREM

17

Problems
1. Prove (5.2.3e). 3. Let : [a, b] be a closed curve in a convex set . Prove that H (t, s) = s (a) + (1 s) (t), t [a, b], s [0, 1] is an -homotopy of to the point (a). 4. Show directly, using the techniques of Problem 3, that a starlike open set is homotopically simply connected. 5. This problem is in preparation for other versions of Runges theorem that appear in Problem 6. Let be an open subset of C, and let {Kn } be as in (5.1.1). Show that in addition to the properties (1), (2) and (3) listed in (5.1.1), the sequence {Kn } has an additional property: \ Kn contains a component of C \ . (4) Each component of C 6. Prove the following versions of Runges theorem: (a) Let be an open set and let S be a set containing at least one point in each \ . Show that if f A(), then there is a sequence {Rn } of rational component of C functions with poles in S such that Rn f uniformly on compact subsets of . (b) Let be an open subset of C. Show that is simply connected if and only if for each f A(), there is a sequence {Pn } of polynomials converging to f uniformly on compact subsets of . 2. Show that h(z ) = z/(1 + |z |) is a homeomorphism of C onto D.

n! + 2 n

1/n

-n!

.
n! 1 + n

. 4/n
Bn Cn

n! +

n!

n!

.7
n

Mn

Kn An

Figure 5.2.2 7. Dene sequences of sets as follows: An = {z : |z + n!| < n! + 2 }, n Bn = {z : |z 4 1 | < }, n n

Ch: 1

123 (5-17)

TOC

+
Index

18

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS Cn = {z : |z (n! + 7 1 )| < n! + }, n n Kn = {z : |z + n!| n! + 7 )| n!} n 1 }, n

4 Ln = { }, n

Mn = {z : |z (n! +

(a) By approximating fn by polynomials (see Problem 6), exhibit a sequence of polynomials converging pointwise to 0 on all of C, but not uniformly on compact subsets. (b) By approximating gn by polynomials, exhibit a sequence of polynomials converging pointwise on all of C to a discontinuous limit.

(see Figure 5.2.2). Dene 0, z An fn (z ) = 1, z Bn 0, z Cn

and gn (z ) =

0, 1,

z An z Cn

5.3

Extending Conformal Maps to the Boundary

Let be a proper simply connected region in C. By the Riemann mapping theorem, there is a one-to-one analytic map of onto the open unit disk D. In this section we will consider the problem of extending f to a homeomorphism of the closure of onto D. Note that if f is extended, then must be compact. Thus we assume in addition that is bounded. We will see that plays an essential role in determining whether such an extension is possible. We begin with some results of a purely topological nature.

5.3.1

Theorem

Suppose is an open subset of C and f is a homeomorphism of onto f () = V . Then a sequence {zn } in has no limit point in i the sequence {f (zn )} has no limit point in V . Proof. Assume {zn } has a limit point z . There is a subsequence {znj } in such that znj z . By continuity, f (znj ) f (z ), and therefore the sequence {f (zn )} has a limit point in V . The converse is proved by applying the preceding argument to f 1 .

5.3.2

Corollary

Proof. Since {zn } has no limit point in , {f (zn )} has no limit point in D, hence |f (zn )| 1.

Suppose f is a conformal equivalence of onto D. If {zn } is a sequence in such that zn , then |f (zn )| 1.

Let us consider the problem of extending a conformal map f to a single boundary point . As the following examples indicate, the relationship of and plays a crucial role.

Ch: 1

124 (5-18)

TOC

Index

5.3. EXTENDING CONFORMAL MAPS TO THE BOUNDARY

19

5.3.3

Examples

(1) Let = C \ (, 0] and let z denote the analytic square root of z such that 1 = 1. Then z is a one-to-one analytic map of onto the right half plane. The linear fractional transformation T (z ) = (z 1)/(z + 1) maps the right half plane onto the unit disk D, hence f (z ) = ( z 1)/( z + 1) is a conformal equivalence of and D. Now T maps Re z = 0 onto D \{1}, so if {zn } is a sequence in Im z > 0 that converges to (, 0), then {f (zn )} converges to a point w D with Im w > 0. On the other hand, if {zn } lies in Im z < 0 and zn , then {f (zn )} converges to a point w D with Im w < 0. Thus f does not have a continuous extension to { } for any on the negative real axis. (2) To get an example of a bounded simply connected region with boundary points to which the mapping functions are not extendible, let = [(0, 1) (0, 1)] \ {{1/n} (0, 1/2] : n = 2, 3, . . . }. Thus is the open unit square with vertical segments of height 1/2 removed at each \ is seen to be of the points 1/2, 1/3, . . . on the real axis; see Figure 5.3.1. Then C

-1

[ , ]
1 2

... iy

.
0 1/5 1/4

.
z2
1/3

. z

1 1/2

Figure 5.3.1 connected, so that is simply connected. Let = iy where 0 < y < 1/2, and choose a sequence {zn } in such that zn iy and Im zn = y, n = 1, 2, 3, . . . . Let f be any conformal map of onto D. Since by (5.3.2), |f (zn )| 1, there is a subsequence {znk } such that {f (znk )} converges to a point w D. For simplicity assume that {f (zn )} converges to w. Set n = f (zn ) and in D, join n to n+1 with the straight line segment [n , n+1 ], n = 1, 2, 3, . . . . Then f 1 ([n , n+1 ]) is a curve in joining zn to zn+1 , n = 1, 2, 3, . . . . It follows that every point of [iy, i/2] is a limit point of n f 1 ([n , n+1 ]). Hence f 1 , in this case, cannot be extended to be continuous at w D. As we now show, if is such that sequences of the type {zn } in the previous example are ruled out, then any mapping function can be extended to { }.

Ch: 1

125 (5-19)

TOC

Index

20

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.3.4

Denition

A point is called simple if to each sequence {zn } in such that zn , there corresponds a curve : [0, 1] { } and a strictly increasing sequence {tn } in [0,1) such that tn 1, (tn ) = zn , and (t) for 0 t < 1.

Thus a boundary point is simple i for any sequence {zn } that converges to , there is a curve in that contains the points zn and terminates at . In Examples 1 and 2 of (5.3.3), none of the boundary points with (, 0) or (0, i/2) is simple.

5.3.5

Theorem

Let be a bounded simply connected region in C, and let be simple. If f is a conformal equivalence of onto D, then f has a continuous extension to { }. To prove this theorem, we will need a lemma due to Lindel of.

. C(z0 ,
z0

r)

Figure 5.3.2

5.3.6

Lemma

Proof. Assume without loss of generality that z0 = 0. Put U = () D(0, r). (This is the shaded region in Figure 5.3.2.) Dene h on U by h(z ) = g (z )g (z ). We claim rst that U () D(0, r). For by general properties of the closure operation, U () D(0, r). Thus it is enough to show that if z D(0, r), that is, |z | = r, / (). But this is a consequence of our assumption that C (0, r) has then z / or z

Suppose is an open set in C, z0 , and the circle C (z0 , r) has an arc lying in the complement of which subtends an angle greater than at z0 (see Figure 5.3.2). Let g on . If |g (z )| M for all z while be any continuous function on which is analytic |g (z )| for all z D(z0 , r) , then |g (z0 )| M .

Ch: 1

126 (5-20)

TOC

Index

5.3. EXTENDING CONFORMAL MAPS TO THE BOUNDARY

21

an arc lying in the complement of that subtends an angle greater than at z0 = 0, from which it follows that the entire circle C (0, r) lies in the complement of (). Consequently, we conclude that if z U , then z D(0, r) or z () D(0, r). Therefore, for all z U , hence for all z U by the maximum principle, we have |h(z )| = |g (z )||g (z )| M. In particular, |h(0)| = |g (0)|2 M , and the lemma is proved. We now proceed to prove Theorem 5.3.5. Assume the statement of the theorem is false. This implies that there is a sequence {zn } in converging to , and distinct complex numbers w1 and w2 of modulus 1, such that f (z2j 1 ) w1 while f (z2j ) w2 . (Proof: There is a sequence {zn } in such that zn while {f (zn )} does not converge. But {f (zn )} is bounded, hence it has at least two convergent subsequences with dierent limits w1 , w2 and with |w1 | = |w2 | = 1.) Let p be the midpoint of the positively oriented arc of D from w1 to w2 . Choose points a and b, interior to this arc, equidistant from p and close enough to p for Figure 5.3.3 to obtain. Let and {tn } be as in the denition

.d W1
w
1

.c .0 W2

. a.

. . p . w b 2
Figure 5.3.3 of simple boundary point. No loss of generality results if we assume that f (z2j 1 ) W1 and f (z2j ) W2 for all j , and that |f ( (t))| > 1/2 for all t. Since f ( (t2j 1 )) W1 and f ( (t2j )) W2 for each j , there exist xj and yj with t2j 1 < xj < yj < t2j such that one of the following holds: (1) f ( (xj )) (0, a), f ( (yj )) (0, b), and f ( (t)) is in the open sector a0ba for all t such that xj < t < yj , or (2) f ( (xj )) (0, d), f ( (yj )) (0, c), and f ( (t)) is in the open sector d0cd for all t such that xj < t < yj . See Figure 5.3.4 for this and details following. Thus (1) holds for innitely many j or (2) holds for innitely many j . Assume that the former is the case, and let J be the set

Ch: 1

127 (5-21)

TOC

Index

22

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

of all j such that (1) is true. For j J dene j on [0, 1] by t 0 t xj , xj f ( (xj )), j (t) = f ( (t)), xj t yj , 1t f ( ( y )) , yj t 1. j 1yj

consists of Thus j is the closed path whose trajectory j

W1
2j -1 ))

(x ) f( j ) ( t

f(

.
a

.
0 j

W2

C(z , r)
0

..z 0

.f( (yj )) . b
w2

.f( (t2j ))

Figure 5.3.4 [0, f ( (xj ))] {f ( (t) : xj t yj } [f ( (yj )), 0].

1 Let j be that component of C \ j such that 2 . Furthermore, p j . Then j j 1 1 j D, for if we compute the index n(j , 2 p), we get 1 because |j (t)| > 2 for xj t yj , while the index of any point in C \ D is 0. Let r be a positive number with r < 1 2 |a b| and choose a point z0 on the open radius (0, p) so close to p that the circle C (0, r) meets the complement of D in an arc of length greater than r. For suciently large j J, |f ( (t))| > |z0 | for all t [t2j 1 , t2j ] ; so for these j we have z0 j . Further, if z j D(z0 , r), then z {f ( (t)) : t2j 1 t t2j } and hence f 1 (z ) ([t2j 1 , t2j ]}. Dene

j = sup{|f 1 (z ) | : z j D(z0 , r)} sup{| (t) | : t [t2j 1 , t2j ]} and M = sup{|f 1 (z ) | : z D}.

Ch: 1

128 (5-22)

TOC

Index

5.3. EXTENDING CONFORMAL MAPS TO THE BOUNDARY

23

Since M sup{|f 1 (z ) | : z j }, Lemma 5.3.6 implies that |f 1 (z0 ) | j M . Since j can be made as small as we please by taking j J suciently large, we have f 1 (z0 ) = . This is a contradiction since f 1 (z0 ) , and the proof is complete.

We next show that if 1 and 2 are simple boundary points and 1 = 2 , then any continuous extension f to {1 , 2 } that results from the previous theorem is one-toone, that is, f (1 ) = f (2 ). The proof requires a lemma that expresses the area of the image of a region under a conformal map as an integral. (Recall that a one-to-one analytic function is conformal.)

5.3.7

Lemma

Let g be a conformal map of an open set . Then the area (Jordan content) of g () is |g |2 dx dy .

Proof. Let g = u + iv and view g as a transformation from R2 into R2 . Since g is analytic, u and v have continuous partial derivatives (of all orders). Also, the Jacobian determinant of the transformation g is (u, v ) = (x, y )
u x v x u y v y

v u u v u v = ( )2 + ( )2 = |g |2 x y x y x x
(u,v ) (x,y )

by the Cauchy-Riemann equations. Since the area of g () is ment of the lemma follows.

dx dy , the state-

f( 2 (t r)) f L
2

.2
2 f( )= 0

f o 2

( r)

. . r .

.. 1

D(1,1)

1 f(

)=

f o 1

g = f -1 f( 1 (s r))
Figure 5.3.5

Ch: 1

129 (5-23)

TOC

Index

24

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.3.8

Theorem

Let be a bounded, simply connected region and f a conformal map of onto D. If 1 and 2 are distinct simple boundary points of and f is extended continuously to {1 , 2 }, then f (1 ) = f (2 ). Proof. Assume that 1 and 2 are simple boundary points of and f (1 ) = f (2 ). We will show that 1 = 2 . It will simplify the notation but result in no loss of generality if we replace D by D(1, 1) and assume that f (1 ) = f (2 ) = 0. Since 1 and 2 are simple boundary points, for j = 1, 2 there are curves j in {j } such that j ([0, 1)) and j (1) = j . Put g = f 1 . By continuity, there exists < 1 such that < s, t < 1 implies |2 (t) 1 (s)| 1 |2 1 | 2 (1)

and there exists , 0 < < 1, such that for t we have f (j (t)) / D(0, ), j = 1, 2. Also, for each r such that 0 < r , we can choose sr and tr > such that f (1 (sr )) and f (2 (tr )) meet the circle C (0, r); see Figure 5.3.5. Let (r) be the principal value of the argument of the point of intersection in the upper half plane of C (0, r) and C (1, 1). Now g (f (2 (tr ))) g (f (1 (sr ))) is the integral of g along the arc r of C (0, r) from f (1 (sr )) to f (2 (tr )). It follows from this and (1) that 1 |2 1 | |2 (tr ) 1 (sr )| 2 = |g (f (2 (tr ))) g (f (1 (sr )))| =|
r (r )

g (z ) dz | |g (rei) |r d. (2) interval need to and the get

(r )

(Note: The function |g (rei )| is positive and continuous on the open ((r), (r)), but is not necessarily bounded. Thus the integral in (2) may be treated as an improper Riemann integral. In any case (2) remains correct calculations that follow are also seen to be valid.) Squaring in (2) and applying the Cauchy-Schwarz inequality for integrals we 1 |2 1 |2 2(r)r2 4
(r ) (r )

|g (rei )|2 d.

(The factor 2(r) comes from integrating 12 d from (r) to (r).) Since (r) /2, we have |2 1 |2 r 4r
(r ) (r )

|g (rei )|2 d.

(3)

Now integrate the right hand side of (3) with respect to r from r = 0 to r = . We obtain
0 (r ) (r )

|g (rei )|2 r d dr

|g (x + iy )|2 dx dy

Ch: 1

130 (5-24)

TOC

Index

5.3. EXTENDING CONFORMAL MAPS TO THE BOUNDARY

25

where L is the lens-shaped open set whose boundary is formed by arcs of C (0, ) and C (1, 1); see Figure 5.3.5. By (5.3.7), |g |2 dx dy is the area (or Jordan content) of L g (L). Since g (L) and is bounded, g (L) has nite area. But the integral from 0 to of the left hand side of (3) is + unless 1 = 2 . Thus f (1 ) = f (2 ) implies that 1 = 2 . We can now prove that f : D extends to a homeomorphism of and D if every boundary point of is simple.

5.3.9

Theorem

Suppose is a bounded, simply connected region with the property that every boundary point of is simple. If f : D is a conformal equivalence, then f extends to a homeomorphism of onto D. Proof. By Theorem 5.3.5, for each we can extend f to { } so that f is continuous on { }. . Assume this has been done. Thus (the extension of) f is a map of into D, and Theorem 5.3.8 implies that f is one-to-one. Furthermore, f is continuous at each point , for if {zn } is any sequence in such that zn then for each n there exists wn with |zn wn | < 1/n and also |f (zn ) f (wn )| < 1/n, by Theorem 5.3.5. But again by (5.3.5), f (wn ) f ( ) because wn and wn . Hence f (zn ) , proving that f is continuous on . Now D f () D, and since f () is compact, hence closed, f () = D. Consequently, f is a one-to-one continuous map of onto D, from which it follows that f 1 is also continuous. Theorem 5.3.9 has various applications, and we will look at a few of these in the sequel. In the proof of (5.3.8), we used the fact that for open subsets L D,
L

|g |2 dx dy

(1)

is precisely the area of g (L), where g is a one-to-one analytic function on D. Suppose that g (z ) = n=0 an z n , z D. Then g (z ) = n=1 nan z n1 . Now in polar coordinates the integral in (1), with L replaced by D, is given by
1 D

|g (rei )|2 r dr d =

r dr
0

|g (rei )|2 d.

But for 0 r < 1,

|g (rei )|2 = g (rei )g (rei )

=
n=1

nan rn1 ei(n1)

m=1

mam rm1 ei(m1)

(2)

=
j =1 m+n=j

nman am rm+n2 ei(nm) .

Since

ei(nm) d =

2, n = m 0, n=m

Ch: 1

131 (5-25)

TOC

Index

26

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

and the series in (2) converges uniformly in , we can integrate term by term to get

|g (re )| d = 2

k=1

k 2 |ak |2 r2k2 .

Multiplying by r and integrating with respect to r, we have

1

r dr
0

|g (rei )|2 d = lim 2

1

k=1

k 2 |ak |2 2k 2k

If this limit, which is the area of g (D), is nite, then

k=1

k |ak |2 < .

We have the following result.

5.3.10

Theorem

Suppose g (z ) = n=0 an z n is one-to-one and analytic on D. If g (D) has nite area, then 2 n=1 n|an | < . Now we will use the preceding result to study the convergence of the power series for g (z ) when |z | = 1. Here is a result on uniform convergence.

5.3.11

Theorem

Let g (z ) = n=0 an z n be a one-to-one analytic map of D onto a bounded region such that every boundary point of is simple. Then the series n=0 an z n converges uniformly on D to (the extension of) g on D.
n Proof. By the maximum principle, it is sucient to show that n=0 an z converges in uniformly to g (z ) for |z | = 1; in other words, n=0 an e converges uniformly in to g (ei ). So let > 0 be given. Since g is uniformly continuous on D, |g (ei ) g (rei )| 0 uniformly in as r 1 . If m is any positive integer and 0 < r < 1, we have m m

|g (ei )

n=0

an ein | |g (ei ) g (rei )| + |g (rei )

n=0

an ein |.

The rst term on the right hand side tends to 0 as r 1 , uniformly in , so let us consider the second term. If k is any positive integer less than m, then since g (rei ) =

Ch: 1

132 (5-26)

TOC

Index

5.3. EXTENDING CONFORMAL MAPS TO THE BOUNDARY

n=0

27

an rn ein , we can write the second term as

k m

n=0

an (rn 1)ein + (1 rn )|an | +

n=k+1 m

an (rn 1)ein +

n=m+1

an rn ein |

n=0 k n=0

n=k+1 m

(1 rn )|an | +

n=m+1

|an |rn |an |rn

n(1 r)|an | +

n=k+1

n(1 r)|an | +

n=m+1

r n1 (since 11 < n). We continue the bounding process by observing r = 1 + r + + r that in the rst of the three above terms, we have n k . In the second term, we write n(1 r)|an | = [ n(1 n |] and apply Schwarzs inequality. In the third term, r)][ n|a we write |an |rn = [ n|an |][rn / n] and again apply Schwarzs inequality. Our bound becomes k m 1/2 m 1/2

k (1 r) +

n=0

|an | +
2

n=k+1 1 /2

n(1 r)

2 n=k+1 1/2

n|an |

n=m+1

n|an |

r 2n n n=m+1

(1)

Since n=0 n|an |2 is convergent, there exists k > 0 such that { n=k+1 n|an |2 }1/2 < /3. Fix such a k . For m > k put rm = (m 1)/m. Now the rst term in (1) is less m than /3 for m suciently large and r = rm . Also, since { n=k+1 n(1 rm )2 }1/2 = m m 2 1 /2 1 /2 { n=k+1 n(1/m) } = (1/m){ n=k+1 n} < (1/m){m(m + 1)/2}1/2 < 1, the middle term in (1) is also less than /3. Finally, consider
2n rm n n=m+1 1/2

rm m+1

2(m+1) 2n rm n=0

1/2

1 rn m + 1 n=0 m

1 /2

which evaluates to 1 1 m + 1 1 rm
1 /2

m m+1

1 /2

< 1.

Thus the last term in (1) is also less than /3 for all suciently large m > k and r = m rm = (m 1)/m. Thus |g (ei ) n=0 an ein | 0 uniformly in as m .

The preceding theorem will be used to produce examples of uniformly convergent power series that are not absolutely convergent. That is, power series that converge uniformly, but to which the Weierstrass M -test does not apply. One additional result will be needed.

Ch: 1

133 (5-27)

TOC

Index

28

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

5.3.12

Theorem
n=0

Suppose f (z ) =

an z n for z D, and
1

n=0

|an | < +. Then for each ,

|f (rei )| dr = lim
1

|f (rei )| dr < +.
n=1

Proof. If 0 r < 1 then for any we have f (rei ) =

0

nan rn1 ei(n1) . Thus

|f (re )| dr

n=1

|an |

n=1

|an | < +.

5.3.13

Remark
1

For each , 0 |f (rei )| dr is the length of the image under f of the radius [0, ei ] of D. For if (r) = rei , 0 r 1, then the length of f is given by
1 0 1 1

|(f ) (r)| dr =

|f (rei )ei | dr =

|f (rei )| dr.

Hence by (5.3.9) and the Riemann mapping theorem (5.2.2), there is a homeomorphism f of D onto such that f is analytic on D. Write f (z ) = an z n , z D. By (5.3.11), i this series converges uniformly on D. Now let e be that point in D such that f (ei ) = 0. Since f is a homeomorphism, f maps the radius of D that terminates at ei onto an arc in {0} that terminates at 0. Further, the image arc in {0} cannot have nite length. Therefore by (5.3.12) we have |an | = +. Additional applications of the results in this section appear in the exercises.

Thus in geometric terms, the conclusion of (5.3.12) is that f maps every radius of D onto an arc of nite length. We can now give a method for constructing uniformly convergent power series that are not absolutely convergent. Let be the bounded, connected, simply connected region that appears in Figure 5.3.6. Then each boundary point of is simple with the possible exception of 0, and the following argument shows that 0 is also a simple boundary point. Let {zn } be any sequence in such that zn 0. For n = 1, 2, . . . put tn = (n1)/n. Then for each n there is a polygonal path n : [tn , tn+1 ] such that n (tn ) = zn , n (tn+1 ) = zn+1 , and such that for tn t tn+1 , Re n (t) is between Re zn and Re zn+1 . If we dene = n , then is a continuous map of [0, 1) into , and (t) = n (t) for tn t tn+1 . Furthermore, (t) 0 as t 1 . Thus by denition, 0 is simple boundary point of .

Problems
1. Let be a bounded simply connected region such that every boundary point of is simple. Prove that the Dirichlet problem is solvable for . That is, if u0 is a real-valued continuous function on , then u0 has a continuous extension u to such that u is harmonic on .

Ch: 1

134 (5-28)

TOC

Index

5.3. EXTENDING CONFORMAL MAPS TO THE BOUNDARY

29

y = x

1/4

1/3

1/2

'

'

'

'

'

y = -x
Figure 5.3.6

2. Let = {x + iy : 0 < x < 1 and x2 < y < x2 }. Show that (a) The identity mapping z z has a continuous argument u on (necessarily harmonic on ). (b) There is a homeomorphism f of D onto which is analytic on D. (c) u f is continuous on D and harmonic on D. (d) No harmonic conjugate V for u f can be bounded on D. 3. Let be a bounded, simply connected region such that every boundary point of is simple. Show that every zero-free continuous function f on has a continuous logarithm g . In addition, show that if f is analytic on , then so is g .

Ch: 1

135 (5-29)

TOC

Index

30

CHAPTER 5. FAMILIES OF ANALYTIC FUNCTIONS

References
1. C. Carath eodory, Conformal Representation, 2nd ed., Cambridge Tracts in Mathematics and Mathematical Physics no. 28, Cambridge University Press, London, 1952. 2. W.P. Novinger, An Elementary Approach to the Problem of Extending Conformal Maps to the Boundary, American Mathematical Monthly, 82(1975), 279-282. 3. W.P. Novinger, Some Theorems from Geometric Function Theory: Applications, American Mathematical Monthly, 82(1975), 507-510. 4. W.A. Veech, A Second Course in Complex Analysis, Benjamin, New York, 1967.

Ch: 1

136 (5-30)

TOC

Index

Chapter 6

Factorization of Analytic Functions

In this chapter we will consider the problems of factoring out the zeros of an analytic function f on a region (` a la polynomials), and of decomposing a meromorphic function (` a la partial fractions for rational functions). Suppose f is analytic on a region and f 0. What can be said about Z (f )? Theorem 2.4.8, the identity theorem, asserts that Z (f ) has no limit point in . It turns out that no more can be said in general. That is, if A is any subset of with no limit point in , then there exists f A() whose set of zeros is precisely A. Furthermore, we can prescribe the order of the zero which f shall have at each point of A. Now if A is a nite subset of , say {z1 , . . . , zn }, and m1 , . . . , mn are the corresponding desired multiplicities, then the nite product f (z ) = (z z1 )m1 (z zn )mn would be such a function. However, in general the construction of such an f is accomplished using innite products, which we now study in detail.

6.1

Innite Products
n

Let {zn } be a sequence of complex numbers and put Pn = k=1 zk , the n-th partial product. We say that the innite product n=1 zn converges if the sequence {Pn } is convergent to a complex number P , and in this case we write P = n=1 zn . This particular denition of convergence of innite products is a natural one if the usual denition of convergence of innite series is extended directly to products. Many textbook authors, however, nd this approach objectionable, primarily for the following two reasons. (a) If one of the factors is zero, then the product converges to zero, no matter what the other factors are, and a correct notion of convergence should presumably depend on all (but possibly nitely many) of the factors. 1
Ch: 1 2 3 4 5 6 7 137 (6-1) TOC Index

CHAPTER 6. FACTORIZATION OF ANALYTIC FUNCTIONS

(b) It is possible for a product to converge to zero without any of the factors being zero, unlike the situation for a nite product. Nevertheless, we have chosen to take the naive approach, and will deal with the above if and when they are relevant. Note that if Pn P = 0 , then zn = Pn /Pn1 P/P = 1 as n . Thus a necessary (but not sucient) condition for convergence of the innite product to a nonzero limit is that zn 1. A natural approach to the study of an innite product is to formally convert the product into a sum by taking logarithms. In fact this approach is quite fruitful, as the next result shows.

6.1.1

Lemma

Suppose that zn = 0, n = 1, 2, . . . . Then n=1 zn converges to a nonzero limit i the series n=1 Log zn converges. (Recall that Log denotes the particular branch of the logarithm such that Im(Log z ) < .) Proof. Let Pn = k=1 zk and Sn = k=1 Log zk . If Sn S , then Pn = eSn eS = 0. Conversely, suppose that Pn P = 0. Choose any such that arg is continuous at P (see Theorem 3.1.2). Then log Pn = ln |Pn | + i arg (Pn ) ln |P | + i arg (P ) = log P . Since eSn = Pn , we have Sn = log Pn + 2iln for some integer ln . But Sn Sn1 = Log zn Log 1 = 0. Consequently, log Pn log Pn1 + 2i(ln ln1 ) 0. Since log Pn log Pn1 log P log P = 0 and ln ln1 is an integer, it follows that ln ln1 is eventually zero. Therefore ln is eventually a constant l. Thus Sn log P + 2il.
n n

6.1.2

Lemma
n=1 (1

If an 0 for all n, then

+ an ) converges i

n=1

an converges.

Proof. Since 1 + x ex , we have, for every n = 1, 2, . . . , a1 + + an (1 + a1 ) (1 + an ) ea1 ++an . Lemma 6.1.2 suggests the following useful notion of absolute convergence for innite products.

6.1.3

Denition

The innite product n=1 (1 + zn ) is said to converge absolutely if n=1 (1 + |zn |) con verges. Thus by (6.1.2), absolute convergence of n=1 (1 + zn ) is equivalent to absolute convergence of the series n=1 zn . With this denition of absolute convergence, we can state and prove a result analogous to a well known property of innite series.

Ch: 1

138 (6-2)

TOC

Index

6.1. INFINITE PRODUCTS

6.1.4

Lemma

If the innite product n=1 (1 + zn ) converges absolutely, then it converges. Proof. By Lemma 6.1.2, convergence of n=1 (1 + |zn |) implies that of n=1 |zn |, hence |zn | 0 in particular. So we can assume that |zn | < 1 for all n. Now for |z | < 1, we have Log(1 + z ) = z where h(z ) = 1
z 2

z2 z3 z4 + + = zh(z ) 2 3 4
p

z2 3

z3 4 p

+ 1 as z 0. Consequently, for m p, Log(1 + zn )| |zn ||h(zn )|.

n=m

n=m

Since {h(zn ) : n = 1, 2, . . . } is a bounded set and n=1 |zn | converges, it follows from the p preceding inequality that | n=m Log(1 + zn )| 0 as m, p . Thus n=1 Log(1 + zn ) is convergent, which by (6.1.1) implies that n=1 (1 + zn ) converges. The preceding result may be combined with (6.1.2) to obtain a rearrangement theorem for absolutely convergent products.

6.1.5

Theorem

If + zn ) converges absolutely, then so does every rearrangement, and to the same limit. That is, if n=1 (1 + |zn |) converges and P = n=1 (1 + zn ), then for every permutation k nk of the positive integers, k=1 (1 + znk ) also converges to P . Proof. Since n=1 (1 + |zn |) converges, so does n=1 |zn | by (6.1.2). But then every rearrangement of this series converges, so by (6.1.2) again, k=1 (1+|znk |) converges. Thus it remains to show that k=1 (1 + znk ) converges to the same limit as does n=1 (1 + zn ). To this end let > 0 and for j = 1, 2, . . . , let Qj be the j -th partial product of k=1 (1 + znk ). Choose N so large that n=N +1 |zn | < and J so large that j J implies that {1, 2, . . . , N } {n1 , n2 , . . . , nj }. (The latter is possible because j nj is a permutation of the positive integers.) Then for j J we have |Qj P | |Qj PN | + |PN P | (1) = |PN ||
k

n=1 (1

(1 + znk ) 1| + |PN P |

where the product is taken over those k j such that nk > N . Now for any complex n n numbers w1 , . . . , wn we have (by induction) | k=1 (1 + wk ) 1| k=1 (1 + |wk |) 1. Using this, we get from (1) that |Qj P | |PN |(
k

(1 + |znk |) 1) + |PN P |

|PN |(e 1) + |PN P |. But the right side of the above inequality can be made as small as we wish by choosing suciently small and N suciently large. Therefore Qj P also, and the proof is complete.

Ch: 1

139 (6-3)

TOC

Index

CHAPTER 6. FACTORIZATION OF ANALYTIC FUNCTIONS

6.1.6

Proposition

Let g1 , g2 , . . . be a sequence of bounded complex-valued functions, each dened on a set S . If the series n=1 |gn | converges uniformly on S , then the product n=1 (1 + gn ) converges absolutely and uniformly on S . Furthermore, if f (z ) = n=1 (1 + gn (z )), z S , then f (z ) = 0 for some z S i 1 + gn (z ) = 0 for some n. Proof. Absolute convergence of the product follows from (6.1.2). If |gn | converges uniformly on S , there exists N such that n N implies |gn (z )| < 1 for all z S . Now for any r N ,
r

(1 + gn (z )) =
n=1

N 1 n=1

(1 + gn (z ))
n=N

(1 + gn (z )).

As in the proof of (6.1.4), with the same h and with m, p N ,

p p

|
n=m

Log(1 + gn (z ))|
n=m

|gn (z )||h(gn (z ))| 0

uniformly on S as m, p . Therefore n=N Log(1 + gn (z )) converges uniformly on S . Since the functions gN , gN +1 , . . . are bounded on S , it follows that the series n=N |gn (z )||h(gn (z ))| is bounded on S and thus by the above inequality, the same is true of n=N Log(1 + gn (z )). However, the exponential function is uniformly continuous on bounded subsets of C, so we may infer that
r

exp
n=N

Log(1 + gn (z ))

exp

n=N

Log(1 + gn (z ))

=0

uniformly on S as r . This proves uniform convergence on S of n=N (1 + gn (z )). Now 1 + gn (z ) is never 0 on S for n N , so if f (z ) = n=1 (1 + gn (z )), then f (z ) = 0 for some z S i 1 + gn (z ) = 0 for some n < N .

Remark
The product
n=1 (1 + |gn |)

also converges uniformly on S , as follows from the inequality

p p

(1 + |gn |) exp
n=m n=m

|gn |

or by applying (6.1.6) to |g1 |, |g2 |, . . . . Proposition (6.1.6) supplies the essential ingredients for an important theorem on products of analytic functions.

6.1.7

Theorem

Let f1 , f2 , . . . be analytic on . If n=1 |fn 1| converges uniformly on compact subsets of , then f (z ) = n=1 fn (z ) denes a function f that is analytic on . Furthermore, for any z we have f (z ) = 0 i fn = 0 for some n.

Ch: 1

140 (6-4)

TOC

Index

6.2. WEIERSTRASS PRODUCTS

Proof. By (6.1.6) with gn = fn 1, the product n=1 fn (z ) converges uniformly on compact subsets of , hence f is analytic on . The last statement of the theorem is also a direct consequence of (6.1.6).

Problems
1. Let f1 , f2 , . . . and f be as in Theorem 6.1.7. Assume in addition that no fn is identically zero on any component of . Prove that for each z , m(f, z ) = n=1 m(fn , z ). (Recall that m(f, z ) is the order of the zero of f at z ; m(f, z ) = 0 if f (z ) = 0.) 2. Show that ln(1 x) = x + g (x)x2 , |x| < 1, where g (x) 1/2 as x 0. Conclude that if a1 , a2 , . . . are real numbers and n=1 an converges, then the innite product 2 2 n (1 an ) converges to a nonzero limit i n=1 an < . Also, if n=1 an < , then n (1 an ) converges to a nonzero limit i n=1 an converges. 3. Determine whether or not the following innite products are convergent. 1)n 1 (b) n (1 n+1 (a) n (1 2n ), ), (c) n (1 + ( ), n (1 n 4. (a) Give an example of an innite product n (1 + an ) diverges. (b) Give an example of an innite product n (1 + an ) converges to a nonzero limit.
n (1 1 n2 ).

+ an ) such that + an ) such that

an converges but an diverges but

n (1

5. Show that the following innite products dene entire functions. (a) n=1 (1 + an z ), |a| < 1, (b) nZ,n=0 (1 z/n)ez/n , (c)
n=2 [1

z n(ln n)2 ].

6. Criticize the following argument. We know that n (1+ zn ) converges to a nonzero limit i n Log(1 + zn ) converges. The Taylor expansion of Log(1 + z ) yields Log(1 + z ) = zg (z ), where g (z ) 1 as z 0. If zn 0, then g (zn ) will be arbitrarily close to 1 for large n, and thus n zn g (zn ) will converge i n zn converges. Consequently, n zn converges. n (1 + zn ) converges to a nonzero limit i

6.2

Weierstrass Products

In this section we will consider the problem of constructing an analytic function f with a prescribed sequence of complex numbers as its set of zeros, as was discussed at the beginning of the chapter. A naive approach is simply to write n (z an )mn where a1 , a2 , . . . is the sequence of (distinct) desired zeros and mn is the specied multiplicity of the zero, that is, m(f, an ) = mn . But if a1 , a2 , . . . is an innite sequence, then the innite product n (z an )mn need not converge. A more subtle approach is required, one that achieves convergence by using factors more elaborate than (z an ). These primary factors were introduced by Weierstrass.

Ch: 1

141 (6-5)

TOC

Index

CHAPTER 6. FACTORIZATION OF ANALYTIC FUNCTIONS

6.2.1

Denition

Dene E0 (z ) = 1 z and for m = 1, 2, . . . , Em (z ) = (1 z ) exp z + z2 zm + + . 2 m

Note that if |z | < 1, then as m , Em (z ) (1 z ) exp[ Log(1 z )] = 1. Indeed, Em (z ) 1 uniformly on compact subsets of the unit disk D. Also, the Em are entire functions, and Em has a zero of order 1 at z = 1, and no other zeros.

6.2.2

Lemma

|1 Em (z )| |z |m+1 for |z | 1. Proof. If m = 0, equality holds, so assume m 1. Then a calculation shows that
Em (z ) = z m exp z +

z2 zm + + 2 m

so that (1 Em (z )) = z m exp z + zm z2 + + . 2 m (1)

This shows that the derivative of 1 Em has a zero of order m at 0. Since 1 Em (0) = 0, it follows that 1 Em has a zero of order m + 1 at z = 0. Thus (1 Em (z ))/z m+1 has a removable singularity at 0 and so has a Taylor expansion n=0 an z n valid everywhere on C. Equation (1) shows also that the derivative of 1 Em has nonnegative Taylor coecients and hence the same must be true of (1 Em (z ))/z m+1 . Thus an 0 for all n. Consequently, 1 Em (z ) an if |z | 1. |an ||z |n z m+1 n=0 n=0 But
n=0

an = [(1 Em (1)]/1m+1 = 1, and the result follows.

Weierstrass primary factors Em will now be used to construct functions with prescribed zeros. We begin by constructing entire functions with given zeros.

6.2.3

Theorem

Let {zn } be a sequence of nonzero complex numbers such that |zn | . Then there is a sequence {mn } of nonnegative integers such that the innite product n=1 Emn (z/zn ) denes an entire function f . Furthermore, f (z ) = 0 i z = zn for some n. Thus it is possible to construct an entire function having zeros precisely at the zn , with prescribed multiplicities. (If a appears k times in the sequence {zn }, then f has a zero of order k at a. Also, a zero at the origin is handled by multiplying the product by z m .)

Ch: 1

142 (6-6)

TOC

Index

6.2. WEIERSTRASS PRODUCTS Proof. Let {mn } be a sequence of nonnegative integers with the property that
n=1

r |zn |

mn +1

<

for every r > 0. (One such sequence is mn = n 1 since for any r > 0, r/|zn |) is eventually less than 1/2.) For xed r > 0, (6.2.2) implies that |1 Emn (z/zn )| |z/zn |mn +1 (r/zn )mn +1 for all z D(0, r). Thus the series |1 Emn (z/zn )| converges uniformly on D(0, r). Since r is arbitrary, the series converges uniformly on compact subsets of C. The result follows from (6.1.7).

6.2.4

Remark

Let {zn } be as in (6.2.3). If |zn | grows suciently rapidly, it may be possible to take {mn } to be a constant sequence. For example, if |zn | = n, then we may choose mn 1. The corresponding product is n=1 E1 (z/zn ) = n=1 (1 z/zn )ez/zn . In this case, m = 1 is the smallest nonnegative integer for which n=1 (r/|zn |)m+1 < for all r > 0, and n=1 Em (z/zn ) can be viewed as the canonical product associated with the sequence {zn }. On the other hand, if |zn | = ln n, then n=1 (1/|zn |)m = + for every nonnegative integer m, so no constant sequence suces. These concepts arise in the study of the order of growth of entire functions, but we will not pursue this area further. Theorem 6.2.3 allows us to factor out the zeros of an entire function.Specically, we have a representation of an entire function as a product involving the primary factors Em .

6.2.5

Weierstrass Factorization Theorem

Let f be an entire function, f 0, and let k 0 be the order of the zero of f at 0. Let the remaining zeros of f be at z1 , z2 , . . . , where each zn is repeated as often as its multiplicity. Then f (z ) = eg(z) z k
n

Emn (z/zn )

for some entire function g and nonnegative integers mn . Proof. If f has nitely many zeros, the result is immediate, so assume that there are innitely many zn . Since f 0, |zn | . By (6.2.3) there is a sequence {mn } such that h(z ) = f (z )/[z k
n=1

Emn (z/zn )]

has a zero-free extension to an entire function, which we will persist in calling h. But now h has an analytic logarithm g on C, hence h(z ) = eg(z) and we have the desired representation. More generally, versions of (6.2.3) and its consequence (6.2.5) are available for any . We begin with the generalization of (6.2.3). proper open subset of C

Ch: 1

143 (6-7)

TOC

Index

CHAPTER 6. FACTORIZATION OF ANALYTIC FUNCTIONS

6.2.6

Theorem

, A = {an : n = 1, 2, . . . } a set of distinct points in Let be a proper open subset of C with no limit point in , and {mn } a sequence of positive integers. Then there exists f A() such that Z (f ) = A and such that for each n we have m(f, an ) = mn . Proof. We rst show that it is sucient to prove the theorem in the special case where and is a deleted neighborhood of in C / A. For suppose that the theorem has been established in this special case. Then let 1 and A1 be arbitrary but as in the hypothesis of the the theorem. Choose a point a = in 1 \ A1 and dene T (z ) = . Then T is a linear fractional transformation of C onto C and thus 1/(z a), z C is a one-to-one continuous map of the open set 1 in C onto an open set . Further, if A = {T (an ) : n = 1, 2 . . . } then and A satisfy the hypotheses of the special case. Having assumed the special case, there exists f analytic on such that Z (f ) = A and m(f, T (an )) = mn . Now consider the function f1 = f T . Since T is analytic on 1 \ {a}, so is f1 . But as z a, T (z ) , and since f is analytic at , f (T (z )) approaches a nonzero limit as z a. Thus f1 has a removable singularity at a with f1 (a) = 0. The statement regarding the zeros of f1 and their multiplicities follows from the fact that T is one-to-one. Now we must establish the special case. First, if A is a nite set {a1 , . . . , an }, then we can simply take f (z ) = (z a1 )m1 (z an )mn (z b)m1 ++mn

where b C \ . The purpose of the denominator is to assure that f is analytic and nonzero at . Now suppose that A = {a1 , a2 , . . . } is an innite set. Let {zn } be a sequence whose range is A but such that for each j , we have zn = aj for exactly mj values of n. Since C \ is a nonempty compact subset of C, for each n 1 there exists a point wn in C \ such that |wn zn | = dist(zn , C \ ). Note that |wn zn | 0 as n because the sequence {zn } has no limit point in . Let {fn } be the sequence of functions on dened by fn (z ) = En zn wn z wn ,

where fn () = En (0) = 1. Then fn has a simple zero at zn and no other zeros. Furthermore, |fn 1| converges uniformly on compact subsets of . For if K , K compact, then eventually |zn wn |/|z wn | is uniformly bounded by 1/2 on K . Thus by Lemma 6.2.2, |fn (z ) 1| = 1 En zn wn z wn zn wn z wn
n+1

(1/2)n+1

for each z K . The statement of the theorem then follows from (6.1.7) by setting f (z ) = n=1 fn (z ).

Ch: 1

144 (6-8)

TOC

Index

6.2. WEIERSTRASS PRODUCTS

It is interesting to see what the preceding argument yields in the special case = C, a case which was established directly in (6.2.3). Specically, suppose that A = {a1 , a2 , . . . } is an innite set of distinct points in C (with no limit point in C), and assume that 0 / A. Let {mj } and {zn } be as in the preceding proof. We are going to reconstruct the proof in the case where \ A. In order to do this, consider the transformation T (z ) = 1/z . \ {0} and the sequence {zn } in C \ {0} onto the sequence {1/zn } This maps C onto C in T (C). The points wn obtained in the proof of (6.2.6) are all 0, and the corresponding functions fn would be given by fn (z ) = En (1/zn z ), z C \ {0}. Thus f (z ) = n=1 fn (z ) is analytic on C \ {0} and f has a zero of order mj at 1/aj . \ {0} back to C, it follows that Transforming C F (z ) = f (1/z ) =
n=1

En (z/zn )

is an entire function with zeros of order mj at aj and no other zeros. That is, we obtain (6.2.3) with mn = n. (Note that this mn from (6.2.3) is unrelated to the sequence {mj } above.) The fact that we can construct analytic functions with prescribed zeros has an interesting consequence, which was referred to earlier in (4.2.5).

6.2.7

Theorem

Let h be meromorphic on the open set C. Then h = f /g where f and g are analytic on . Proof. Let A be the set of poles of h in . Then A satises the hypothesis in (6.2.6). Let g be an analytic function on with zeros precisely at the points in A and such that for each a A, the order of the zero of g at a equals the order of the pole of h at a. Then gh has only removable singularities in and thus can be extended to an analytic function f A().

Problems
1. Determine the canonical products associated with each of the following sequences. [See the discussion in (6.2.4).] (a) zn = 2n , (b) zn = nb , b > 0, (c) zn = n(ln n)2 . 2. Apply Theorem 6.2.6 to construct an analytic function f on the unit disk D such that f has no proper analytic extension to a region D. (Hint: Construct a countable set A = {an : n = 1, 2, . . . } in D such that every point in D is an accumulation point of A.) Compare this approach to that in Theorem 4.9.5, where essentially the same result is obtained by quite dierent means.

Ch: 1

145 (6-9)

TOC

Index

10

CHAPTER 6. FACTORIZATION OF ANALYTIC FUNCTIONS

6.3

Mittag-Leers Theorem and Applications

Let be an open subset of C and let A = {an : n = 1, 2, . . . } be a set of distinct points in with no limit point in . If {mn } is a sequence of positive integers, then Theorem 6.2.6 implies (by using 1/f ) that there is a meromorphic function f on such that f has poles of order precisely mn at precisely the points an . The theorem of Mittag-Leer, which we will prove next, states that we can actually specify the coecients of the principal part at each pole an . The exact statement follows; the proof requires Runges theorem.

6.3.1

Mittag-Leers Theorem

Let be an open subset of C and B a subset of with no limit point in . Thus B = {bj : j J } where J is some nite or countably innite index set. Suppose that to each j J there corresponds a rational function of the form Sj (z ) = ajnj aj 1 aj 2 + . + + 2 z bj (z bj ) (z bj )nj

Then there is a meromorphic function f on such that f has poles at precisely the points bj and such that the principal part of the Laurent expansion of f at bj is exactly Sj . Proof. Let {Kn } be the sequence of compact sets dened in (5.1.1). Recall that {Kn } o has the properties that Kn Kn +1 and Kn = . Furthermore, by Problem 5.2.5, each component of C \ Kn contains a component of C \ , in particular, C \ meets each component of C \ Kn . Put K0 = and for n = 1, 2, . . . , dene Jn = {j J : bj Kn \ Kn1 }. The sets Jn are pairwise disjoint (possibly empty), each Jn is nite (since B has no limit point in ), and Jn = J . For each n, dene Qn by Qn (z ) =
j Jn

Sj (z )

where Qn 0 if Jn is empty. Then Qn is a rational function whose poles lie in Kn \ Kn1 . In particular, Qn is analytic on a neighborhood of Kn1 . Hence by Runges theorem (5.2.8) with S = C \ , there is a rational function Rn whose poles lie in C \ such that |Qn (z ) Rn (z )| (1/2)n , z Kn1 . It follows that for any xed m 1, the series n=m+1 (Qn Rn ) converges uniformly on o Km to a function which is analytic on Km Km1 . Thus it is meaningful to dene a function f : C by f (z ) = Q1 (z ) +
n=2

(Qn (z ) Rn (z )), z .
m

Indeed, note that for any xed m, f is the sum of the rational function Q1 + n=2 (Qn Rn ) o and the series n=m+1 (Qn Rn ), which is analytic on Km . Therefore f is meromorphic

Ch: 1

146 (6-10)

TOC

Index

6.3. MITTAG-LEFFLERS THEOREM AND APPLICATIONS

11

on , as well as analytic on \ B . It remains to show that f has the required principal part at each point b B . But for any bj B , we have f (z ) = Sj (z ) plus a function that is analytic on a neighborhood of bj . Thus f has a pole at bj with the required principal part Sj .

6.3.2

Remark

Suppose g is analytic at the complex number b and g has a zero of order m 1 at b. Let c1 , c2 , . . . , cm be given complex numbers, and let R be the rational function given by R(z ) = c1 cm + + . zb (z b)m

Then gR has a removable singularity at b, so there exist complex numbers a0 , a1 , a2 , . . . such that for z in a neighborhood of b, g (z )R(z ) = a0 + a1 (z b) + + am1 (z b)m1 + . Furthermore, if we write the Taylor series expansion g (z ) = b0 (z b)m + b1 (z b)m+1 + + bm1 (z b)2m1 + , then the coecients a0 , a1 , . . . for gR must satisfy a0 = b0 cm a1 = b0 cm1 + b1 cm . . . am1 = b0 c1 + b1 c2 + + bm1 cm That is, if c1 , c2 , . . . , cm are given, then a0 , a1 , . . . , am1 are determined by the above equations. Conversely, if g is given as above, and a0 , a1 , . . . , am1 are given complex numbers, then since b0 = 0, one can sequentially solve the equations to obtain, in order, cm , cm1 , . . . , c1 . This observation plays a key role in the next result, where it is shown that not only is it possible to construct analytic functions with prescribed zeros and with prescribed orders at these zeros, as in (6.2.3) and (6.2.6), but we can specify the values of f and nitely many of its derivatives in an arbitrary way. To be precise, we have the following extension of (6.2.6).

6.3.3

Theorem

Let be an open subset of C and B a subset of with no limit point in . Index B by J , as in Mittag-Leers theorem, so B = {bj : j J }. Suppose that corresponding to each j J , there is a nonnegative integer nj and complex numbers a0j , a1j , . . . , anj ,j . Then there exists f A() such that for each j J , f (k) (bj ) = akj , 0 k nj . k!

Ch: 1

147 (6-11)

TOC

Index

12

CHAPTER 6. FACTORIZATION OF ANALYTIC FUNCTIONS

Proof. First apply (6.2.6) to produce a function g A() such that Z (g ) = B and for each j, m(g, bj ) = nj + 1 = mj , say. Next apply the observations made above in (6.3.2) to obtain, for each bj B , complex numbers c1j , c2j , . . . , cmj ,j such that
mj

g (z )
k=1

ckj = a0j + a1j (z bj ) + + anj ,j (z bj )nj + (z bj )k

for z near bj . Finally, apply Mittag-Leers theorem to obtain h, meromorphic on , such that for each j ,
mj

h
k=1

ckj (z bj )k

has a removable singularity at bj . It follows that the analytic extension of gh to is the required function f . (To see this, note that
mj

gh = g h
k=1

ckj (z bj )k

mj

+g
k=1

ckj (z bj )k

and m(g, bj ) > nj .)

6.3.4

Remark

Theorem 6.3.3 will be used to obtain a number of algebraic properties of the ring A(). This theorem, together with most of results to follow, were obtained (in the case = C) by Olaf Helmer, Duke Mathematical Journal, volume 6, 1940, pp.345-356. Assume in what follows that is connected. Thus by Problem 2.4.11, A() is an integral domain. Recall that in a ring, such as A(), g divides f if f = gq for some q A(). Also, g is a greatest common divisor of a set F if g is a divisor of each f F and if h divides each f F , then h divides g .

6.3.5

Proposition

Each nonempty subfamily F A() has a greatest common divisor, provided F = {0}. Proof. Put B = {Z (f ) : f F}. Apply Theorem 6.2.6 to obtain g A() such that Z (g ) = B and for each b B, m(g, b) = min{m(f, b) : f F}. Then f F implies that g |f (g divides f ). Furthermore, if h A() and h|f for each f F , then Z (h) B and for each b B, m(h, b) min{m(f, b) : f F} = m(g, b). Thus h|g , and consequently g is a greatest common divisor of F .

6.3.6

Denitions

A unit in A() is a function f A() such that 1/f A(). Thus f is a unit i f has no zeros in . If f, g A(), we say that f and g are relatively prime if each greatest common divisor of f and g is a unit. It follows that f and g are relatively prime i Z (f ) Z (g ) = . (Note that f and g have a common zero i they have a nonunit common factor.)

Ch: 1

148 (6-12)

TOC

Index

6.3. MITTAG-LEFFLERS THEOREM AND APPLICATIONS

13

6.3.7

Proposition

If the functions f1 , f2 A() are relatively prime, then there exist g1 , g2 A() such that f1 g1 + f2 g2 1. Proof. By the remarks above, Z (f1 ) Z (f2 ) = . By working backwards, i.e., solving f1 g1 + f2 g2 = 1 for g1 , we see that it suces to obtain g2 such that (1 f2 g2 )/f1 has only removable singularities. But this entails obtaining g2 such that Z (f1 ) Z (1 f2 g2 ) and such that for each a Z (f1 ), m(f1 , a) m(1 f2 g2 , a). However, the latter condition may be satised by invoking (6.3.3) to obtain g2 A() such that for each a Z (f1 ) (recalling that f2 (a) = 0), 0 = 1 f2 (a)g2 (a) = (1 f2 g2 )(a)
(a) + f2 (a)g2 (a) = (1 f2 g2 ) (a) 0 = f2 (a)g2 0 = f2 (a)g2 (a) + 2f2 (a)g2 (a) + f2 (a)g2 (a) = (1 f2 g2 ) (a) . . .

0 = f2 (a)g2

(m1)

(a) + + f2

(m1)

(a)g2 (a) = (1 f2 g2 )(m1) (a)

(m1)

where m = m(f1 , a). [Note that these equations successively determine g2 (a), g2 (a), . . . , g2 This completes the proof of the proposition.

(a).]

The preceding result can be generalized to an arbitrary nite collection of functions.

6.3.8

Proposition

If {f1 , f2 , . . . , fn } A() and d is a greatest common divisor for this set, then there exist g1 , g2 , . . . , gn A() such that f1 g1 + f2 g2 + + fn gn = d. Proof. Use (6.3.7) and induction. The details are left as an exercise (Problem 1). Recall that an ideal I A() is a subset that is closed under addition and subtraction and has the property that if f A() and g I , then f g I . We are now going to show that A() is what is referred to in the literature as a Bezout domain. This means that each nitely generated ideal in the integral domain A() is a principal ideal. A nitely generated ideal is an ideal of the form {f1 g1 + + fn gn : g1 , . . . , gn A()} where {f1 , . . . , fn } is some xed nite set of elements in A(). A principal ideal is an ideal that is generated by a single element f1 . Most of the work has already been done in preceding two propositions.

6.3.9

Theorem

Let f1 , . . . , fn A() and let I = {f1 g1 + + fn gn : g1 , . . . , gn A()} be the ideal generated by f1 , . . . , fn . Then there exists f A() such that I = {f g : g A()}. In other words, I is a principal ideal. Proof. If f I then f = f1 h1 + + fn hn for some h1 , . . . , hn A(). If d is a greatest common divisor for {f1 , . . . , fn }, then d divides each fj , hence d divides f . Thus f is a multiple of d. On the other hand, by (6.3.8), there exist g1 , . . . , gn A() such that

Ch: 1

149 (6-13)

TOC

Index

14

CHAPTER 6. FACTORIZATION OF ANALYTIC FUNCTIONS

d = f1 g1 + + fn gn . Therefore d and hence every multiple of d belongs to I . Thus I is the ideal generated by the single element d. A principal ideal domain is an integral domain in which every ideal is principal. Problem 2 asks you to show that A() is never a principal ideal domain, regardless of the region . There is another class of (commutative) rings called Noetherian ; these are rings in which every ideal is nitely generated. Problem 2, when combined with (6.3.9), also shows that A() is never Noetherian.

Problems
1. Supply the details to the proof of (6.3.8). (Hint: Use induction, (6.3.7), and the fact that if d is a greatest common divisor (gcd) for {f1 , . . . , fn } and d1 is a gcd for {f1 , . . . , fn1 }, then d is a gcd for the set {d1 , fn }. Also note that 1 is a gcd for {f1 /d, . . . , fn /d}.) 2. Show that A() is never a principal ideal domain. that is, there always exists ideals I that are not principal ideals, and thus by (6.3.9) are not nitely generated. (Hint: Let {an } be a sequence of distinct points in with no limit point in . For each n, apply (6.2.6) to the set {an , an+1 , . . . }.)

Ch: 1

150 (6-14)

TOC

Index

Chapter 7

The Prime Number Theorem

In this nal chapter we will take advantage of an opportunity to apply many of the ideas and results from earlier chapters in order to give an analytic proof of the famous prime number theorem: If (x) is the number of primes less than or equal to x, then x1 (x) ln x 1 as x . That is, (x) is asymptotically equal to x/ ln x as x . (In the sequel, prime will be taken to mean positive prime.) Perhaps the rst recorded property of (x) is that (x) as x , in other words, the number of primes is innite. This appears in Euclids Elements. A more precise result that was established much later by Euler (1737) is that the series of reciprocals of the prime numbers, 1 1 1 1 1 + + + + + , 2 3 5 7 11 is a divergent series. This can be interpreted in a certain sense as a statement about how fast (x) as x . Later, near the end of the 18-th century, mathematicians, including Gauss and Legendre, through mainly empirical considerations, put forth conjectures that are equivalent to the above statement of the prime number theorem (PNT). However, it was not until nearly 100 years later, after much eort by numerous 19-th century mathematicians, that the theorem was nally established (independently) by Hadamard and de la Vall ee Poussin in 1896. The quest for a proof led Riemann, for example, to develop complex variable methods to attack the PNT and related questions. In the process, he made a remarkable and as yet unresolved conjecture known as the Riemann hypothesis, whose precise statement will be given later. Now it is not clear on the surface that there is a connection between complex analysis and the distribution of prime numbers. But in fact, every proof of the PNT dating from Hadamard and de la Vall ee Poussin, up to 1949 when P. Erd os and A.Selberg succeeded in nding elementary proofs, has used the methods of complex variables in an essential way. In 1980, D.J. Newman published a new proof of the PNT which, although still using complex analysis, nevertheless represents a signicant simplication of previous proofs. It is Newmans proof, as modied by J. Korevaar, that we present in this chapter. There are a number of preliminaries that must be dealt with before Newmans method can be applied to produce the theorem. The proof remains far from trivial but the steps 1
Ch: 1 2 3 4 5 6 7 151 (7-1) TOC Index

CHAPTER 7. THE PRIME NUMBER THEOREM

along the way are of great interest and importance in themselves. We begin by introducing the Riemann zeta function, which arises via Eulers product formula and forms a key link between the sequence of prime numbers and the methods of complex variables.

7.1

The Riemann Zeta function

1 z n n=1

The Riemann zeta function is dened by (z ) =

where nz = ez ln n . Since |nz | = nRe z , the given series converges absolutely on Re z > 1 and uniformly on {z : Re z 1 + } for every > 0. Let p1 , p2 , p3 , . . . be the sequence 2,3,5, . . . of prime numbers and note that for j = 1, 2, . . . and Re z > 1, we have 1 1 1 = 1 + z + 2z + . 1 1/pz p p j j j Now consider the partial product 1 1 1 (1 + z + 2z + ). z = p p 1 p j j j j =1 j =1 By multiplying the nitely many absolutely convergent series on the right together, rearranging, and applying the fundamental theorem of arithmetic, we nd that the product is 1 the same as the sum nPm n z , where Pm consists of 1 along with those positive integers whose prime factorization uses only primes from the set {p1 , . . . , pm }. Therefore 1 1 , Re z > 1. z = z n 1 p j n=1 j =1 We now state this formally.
m m

7.1.1

Eulers Product formula

n=1

For Re z > 1, the Riemann zeta function (z ) =

j =1

1/nz is given by the product

1 z 1 p j

where {pj } is the (increasing) sequence of prime numbers. The above series and product converge uniformly on compact subsets of Re z > 1, hence is analytic on Re z > 1. Furthermore, the product representation of shows that has no zeros in Re z > 1 (Theorem 6.1.7). Our proof of the PNT requires a number of additional properties of . The rst result is concerned with extending to a region larger than Re z > 1.

Ch: 1

152 (7-2)

TOC

Index

7.1. THE RIEMANN ZETA FUNCTION

7.1.2

Extension Theorem for Zeta

The function (z ) 1/(z 1) has an analytic extension to the right half plane Re z > 0. Thus has an analytic extension to {z : Re z > 0, z = 1} and has a simple pole with residue 1 at z = 1. Proof. For Re z > 1, apply the summation by parts formula (Problem 2.2.7) with an = n and bn = 1/nz to obtain
k1 n=1

1 1 z z (n + 1) n

1 k z1

1 . (n + 1)z n=1

k1

Thus 1 1 1 1 1+ n = z1 z . z z ( n + 1) k ( n + 1) n n=1 n=1 But n 1 1 z (n + 1)z n

n+1 n+1 k1 k1

= nz

tz1 dt = z

[t]tz1 dt
n

where [t] is the largest integer less than or equal to t. Hence we have 1 1 1 =1+ = z1 + z z z n (n + 1) k n=1 n=1 n=1 = 1 k z1
k k k1 k1 n+1

[t]tz1 dt
n

+z
1

[t]tz1 dt.

Letting k , we obtain the integral formula (z ) = z

1

[t]tz1 dt

(1)

for Re z > 1. Consider, however, the closely related integral z

1

ttz1 dt = z
1

tz dt =

z 1 =1+ . z1 z1

Combining this with (1) we can write (z ) 1 =1+z z1

k 1

([t] t)tz1 dt.

Now x k > 1 and consider the integral 1 ([t] t)tz1 dt. By (3.3.3), this integral is an entire function of z . furthermore, if Re z > 0, then
k k

([t] t)tz1 dt|

t Re(z+1) dt

t1Re z dt =

1 . Re z

Ch: 1

153 (7-3)

TOC

Index

CHAPTER 7. THE PRIME NUMBER THEOREM

k

This implies that the sequence fk (z ) = 1 ([t] t)tz1 dt of analytic functions on Re z > 0 is uniformly bounded on compact subsets. Hence by Vitalis theorem (5.1.14), the limit function f (z ) =
1

([t] t)tz1 dt

(as the uniform limit on compact subsets of Re z > 0) is analytic, and thus the function 1+z
1

([t] t)tz1 dt

1 is also analytic on Re z > 0. But this function agrees with (z ) z 1 for Re z > 1, and consequently provides the required analytic extension of to Re z > 0. This completes the proof of the theorem. We have seen that Eulers formula (7.1.1) implies that has no zeros in the half plane Re z > 1, but how about zeros of (the extension of) in 0 < Re z 1? The next theorem asserts that has no zeros on the line Re z = 1. This fact is crucial to our proof of the PNT.

7.1.3

Theorem

The Riemann zeta function has no zeros on Re z = 1, so (z 1) (z ) is analytic and zero-free on a neighborhood of Re z 1. Proof. Fix a real number y = 0 and consider the auxiliary function h(x) = 3 (x) 4 (x + iy ) (x + i2y ) for x real and x > 1. By Eulers product formula, if Re z > 1 then ln | (z )| =
j =1 z ln |1 p j | = Re j =1 z Log(1 p j ) = Re n=1

1 nz p n j j =1 n=1

where we have used the expansion Log(1 w) =

wn /n, valid for |w| < 1. Hence

ln |h(x)| = 3 ln | (x)| + 4 ln | (x + iy )| + ln | (x + i2y )| 1 nxiny 1 nx pj + 4 Re pj = 3 Re n n j =1 n=1 j =1 n=1 + Re =

1 nxi2ny p n j j =1 n=1

1 nx iny i2ny pj Re(3 + 4p + p ). j j n j =1 n=1

iny i2ny iny i2ny But p = einy ln pj and p = ei2ny ln pj . Thus Re(3 + 4p + pj ) has the j j j form

3 + 4 cos + cos 2 = 3 + 4 cos + 2 cos2 1 = 2(1 + cos )2 0.

Ch: 1

154 (7-4)

TOC

Index

7.1. THE RIEMANN ZETA FUNCTION Therefore ln |h(x)| 0 and consequently |h(x)| = | 3 (x)|| 4 (x + iy )|| (x + i2y )| 1. Thus (x + iy ) |h(x)| = |(x 1) (x)|3 x1 x1
4

| (x + i2y )|

1 . x1

But if (1 + iy ) = 0, then the left hand side of this inequality would approach a nite limit | (1 + iy )|4 | (1 + i2y )| as x 1+ since has a simple pole at 1 with residue 1. However, the right hand side of the inequality contradicts this. We conclude that (1 + iy ) = 0. Since y is an arbitrary nonzero real number, has no zeros on Re z = 1.

Remark
The ingenious introduction of the auxiliary function h is due to Mertens (1898). We now have shown that any zeros of in Re z > 0 must lie in the strip 0 < Re z < 1. The study of the zeros of has long been the subject of intensive investigation by many mathematicians. Riemann had stated in his seminal 1859 paper that he considered it very likely that all the zeros of in the above strip, called the critical strip, lie on the line Re z = 1/2. This assertion is now known as the Riemann hypothesis, and remains as yet unresolved. However, a great deal is known about the distribution of the zeros of in the critical strip, and the subject continues to capture the attention of eminent mathematicians. To state just one such result, G.H. Hardy proved in 1915 that has innitely many zeros on the line Re z = 1/2. Those interested in learning more about this fascinating subject may consult, for example, the book Riemanns Zeta Function by H.M. Edwards. Another source is http://mathworld.wolfram.com/RiemannHypothesis.html. We turn next to zetas logarithmic derivative / , which we know is analytic on Re z > 1. In fact, more is true, for by (7.1.3), / is analytic on a neighborhood of {z : Re z 1 and z = 1}. Since has a simple pole at z = 1, so does / , with residue Res( /, 1) = 1. [See the proof of (4.2.7).] We next obtain an integral representation for / that is similar to the representation (1) above for . [See the proof of (7.1.2).] But rst, we must introduce the von Mangoldt function , which is dened by (n) = ln p 0 if n = pm for some m, otherwise.

Thus (n) is ln p if n is a power of the prime p, and is 0 if not. Next dene on x 0 by (x) =
nx

(n).

(2)

An equivalent expression for is (x) =

px

mp (x) ln p,

Ch: 1

155 (7-5)

TOC

Index

CHAPTER 7. THE PRIME NUMBER THEOREM

where the sum is over primes p x and mp (x) is the largest integer such that pmp (x) x. (For example, (10.4) = 3 ln 2 + 2 ln 3 + ln 5 + ln 7.) Note that pmp (x) x i mp (x) ln p

ln x x ln x i mp (x) ln ln p . Thus mp (x) = ln p where as before, [ ] denotes the greatest integer function. The function will be used to obtain the desired integral representation for / .

7.1.4

Theorem
(z ) =z (z )
1

For Re z > 1, (t)tz1 dt (3)

where is dened as above. Proof. In the formulas below, p and q range over primes. If Re z > 1, we have (z ) = z 1 ) by (7.1.1), hence p (1 p (z ) =
p

pz ln p (1 pz )2
z p

q =p

1 1 q z

= (z ) = (z )
p

p ln p (1 pz ) (1 pz )2 pz ln p . 1 pz
p n=1

Thus (z ) = (z ) pz ln p = 1 pz pnz ln p.

The iterated sum is absolutely convergent for Re z > 1, so it can be rearranged as a double sum (pn )z ln p =
(p,n),n1 k

k z ln p

where k = pn for some n. Consequently, (z ) = (z )

k=1

k z (k ) =

k=1

k z ( (k ) (k 1))

by the denitions of and . But using partial summation once again we obtain, with ak = k z , bk+1 = (k ), and b1 = (0) = 0 in Problem 2.2.7,
M M

k=1

k z ( (k ) (k 1)) = (M )(M + 1)z +

k=1

(k )(k z (k + 1)z ).

Ch: 1

156 (7-6)

TOC

Index

7.2. AN EQUIVALENT VERSION OF THE PRIME NUMBER THEOREM

Now from the denition (2) of (x) we have (x) x ln x, so if Re z > 1 we have (M )(M + 1)z 0 as M . Moreover, we can write
M M k+1

k=1

(k )(k z (k + 1)z ) = =

(k )z
k=1 M k+1 k

tz1 dt (t)tz1 dt

z
k=1 M k

=z
1

(t)tz1 dt

because is constant on each interval [k, k + 1). Taking limits as M , we nally get (z ) =z (z )
1

(t)tz1 dt, Re z > 1.

7.2

An Equivalent Version of the Prime Number Theorem

The function dened in (2) above provides yet another connection, through (3), between the Riemann zeta function and properties of the prime numbers. The integral that appears in (3) is called the Mellin transform of and is studied in its own right. We next establish a reduction, due to Chebyshev, of the prime number theorem to a statement involving the function .

7.2.1

Theorem

The prime number theorem holds, that is, x1 (x) ln x 1, i x1 (x) 1 as x . Proof. Recall that (x) =
px

ln x ln p ln p ln x ln p ln p 1
px

(1)

px

= ln x

= (ln x) (x).

Ch: 1

157 (7-7)

TOC

Index

8 However, if 1 < y < x, then

CHAPTER 7. THE PRIME NUMBER THEOREM

(x) = (y ) +
y<px

1 ln p ln y ln p
y<px

(y ) + <y+ y+ Now take y = x/(ln x)2 in (2), and we get (x) Thus (x)

(2)

1 ln y

y<px

1 (x). ln y

1 x + (x). (ln x)2 ln x 2 ln ln x

ln x 1 ln x (x) + . x ln x ln x 2 ln ln x x

(3)

It now follows from (1) and (3) that (x) ln x 1 ln x (x) (x) + x x ln x ln x 2 ln ln x x and from this we can see that x1 (x) 1 i x1 (x) ln x 1 as x . The goal will now be to show that (x)/x 1 as x . A necessary intermediate step for our proof is to establish the following weaker estimate on the asymptotic behavior of (x).

7.2.2

Lemma

There exists C > 0 such that (x) Cx, x > 0. For short, (x) = O(x). x Proof. Again recall that (x) = px [ ln ln p ] ln p, x > 0. Fix x > 0 and let m be an integer such that 2m < x 2m+1 . Then (x) = (2m ) + (x) (2m ) =
p2m

(2m ) + (2m+1 ) (2m ) ln 2m ln p + ln p

2m <p2m+1

ln 2m+1 ln p. ln p

Consider, for any positive integer n, ln p = ln

n<p2n n<p2n

p.

Ch: 1

158 (7-8)

TOC

Index

7.3. PROOF OF THE PRIME NUMBER THEOREM Now for any prime p such that n < p 2n, p divides (2n)!/n! = n! n does not divide n!, it follows that p divides 2n . Hence p 2n n < (1 + 1)2n = 22n ,
2n n

9 . Since such a p

n<p2n

and we arrive at ln p < 2n ln 2.

n<p2n

Therefore
m

ln p =
p2m k=1

2k1 <p2k

and

ln p <

2k ln 2 < 2m+1 ln 2
k=1

ln p < 2m+1 ln 2.
2m <p2m+1

But if p x is such that

x p. x ln x Thus those terms in the sum px ln x, ln p ln p where ln p > 1 occur only when p and the sum of terms of this form contribute no more than
p x

ln x ln p

> 1, then

ln x ln p

2 and hence x p2 so that

ln x ln p = ( x) ln x. ln p

It follows from the above discussion that if 2m < x 2m+1 , then (x) 2m+1 ln 2 + 2m+1 ln 2 + ( x) ln x = 2m+2 ln 2 + ( x) ln x < 4x ln 2 + ( x) ln x 4x ln 2 + x ln x 1 = (4 ln 2 + ln x)x. x Since
1 x

ln x 0 as x , we conclude that (x) = O(x), which proves the lemma.

7.3

Proof of the Prime Number Theorem

Our approach to the prime number theorem has been along traditional lines, but at this stage we will apply D.J. Newmans method (Simple Analytic Proof of the Prime Number Theorem, American Math. Monthly 87 (1980), 693-696) as modied by J. Korevaar (On

Ch: 1

159 (7-9)

TOC

Index

10

CHAPTER 7. THE PRIME NUMBER THEOREM

Newmans Quick Way to the Prime Number Theorem, Math. Intelligencer 4 (1982), 108115). Korevaars approach is to apply Newmans ideas to obtain properties of certain Laplace integrals that lead to the prime number theorem. Our plan is to deduce the prime number theorem from a Tauberian theorem (7.3.1) and its corollary (7.3.2). Then we will prove (7.3.1) and (7.3.2).

7.3.1

Auxiliary Tauberian Theorem

Let F be bounded and piecewise continuous on [0, +), so that its Laplace transform G(z ) =
0

F (t)ezt dt

exists and is analytic on Re z > 0. Assume that G has an analytic extension to a neigh borhood of the imaginary axis, Re z = 0. Then 0 F (t) dt exists as an improper integral and is equal to G(0). [In fact, 0 F (t)eiyt dt converges for every y R to G(iy ).] Results like (7.3.1) are named for A. Tauber, who is credited with proving the rst theorem of this type near the end of the 19th century. The phrase Tauberian theorem was coined by G.H. Hardy, who along with J.E. Littlewood made a number of contributions in this area. Generally, Tauberian theorems are those in which some type of ordinary convergence (e.g., convergence of 0 F (t)eiyt dt for each y R), is deduced from some weaker type of convergence (e.g., convergence of 0 F (t)ezt dt for each z with Re z > 0) provided additional conditions are satised (e.g., G has an analytic extension to a neighborhood of each point on the imaginary axis). Taubers original theorem can be found in The Elements of Real Analysis by R.G. Bartle.

7.3.2

Corollary

Let f be a nonnegative, piecewise continuous and nondecreasing function on [1, ) such that f (x) = O(x). Then its Mellin transform g (z ) = z
1

f (x)xz1 dx

has an analytic extension to a neighborhood of the line Re z = 1. Then as x , f (x) c. x

exists for Re z > 1 and denes an analytic function g . Assume that for some constant c, the function c g (z ) z1

As stated earlier, we are rst going to see how the prime number theorem follows from (7.3.1) and (7.3.2). To this end, let be as above, namely (x) =
px

ln x ln p. ln p

Ch: 1

160 (7-10)

TOC

Index

7.3. PROOF OF THE PRIME NUMBER THEOREM

11

Then is a nonnegative, piecewise continuous, nondecreasing function on [1, ). Furthermore, by (7.2.2), (x) = O(x), so by (7.3.2) we may take f = and consider the Mellin transform g (z ) = z
1

(x)xz1 dx.

But by (7.1.4), actually g (z ) = (z )/ (z ), and by the discussion leading up to the (z ) 1 statement of (7.1.4), (z ) + z 1 has an analytic extension to a neighborhood of each 1 point of Re z = 1, hence so does g (z ) z 1 . Consequently, by (7.3.2), we can conclude that (x)/x 1, which, by (7.2.1), is equivalent to the PNT. Thus we are left with the proof of (7.3.1) and its corollary (7.3.2).

Proof of (7.3.1)
Let F be as in the statement of the theorem. Then it follows just as in the proof of (7.1.2), the extension theorem for zeta, that F s Laplace transform G is dened and analytic on Re z > 0. Assume that G has been extended to an analytic function on a region containing Re z 0. Since F is bounded we may as well assume that |F (t)| 1, t 0. For 0 < < , dene

G (z ) =
0

F (t)ezt dt.

By (3.3.3), each function G is entire, and the conclusion of our theorem may be expressed as

lim G (0) = G(0).

That is, the improper integral 0 F (t) dt exists and converges to G(0). We begin the analysis by using Cauchys integral formula to get a preliminary estimate of |G (0) G(0)|. For each R > 0, let (R) > 0 be so small that G is analytic inside and on the closed path

.iR
R
R) (

+ R

. .

+ R = R +

.-iR
Figure 7.3.1 R in Figure 7.3.1. (Note that since G is analytic on an open set containing Re z 0,

Ch: 1

161 (7-11)

TOC

Index

12

CHAPTER 7. THE PRIME NUMBER THEOREM

such a (R) > 0 must exist, although it may well be the case that (R) 0 as R +.) + the portion that lies in denote that portion of R that lies in Re z > 0, and R Let R Re z < 0. By Cauchys integral formula, G(0) G (0) = 1 2i 1 (G(z ) G (z )) dz. z R (1)

Let us consider the consequences of estimating |G(0) G (0)| by applying the usual M+ and L estimates to the integral on the right hand side of (1) above. First, for z R x = Re z , we have
1 G(z ) G (z ) = F (t)ezt dt z R 1 |F (t)|ext dt R 1 xt e dt R 1 ex = R x 1 1 1 1 = . Rx R Re z

(2)

+ , so we see that a more delicate approach is required to But 1/ Re z is unbounded on R shows that G(0) G (0) 0 as . Indeed, it is here that Newmans ingenuity comes to the fore, and provides us with a modication of the above integral representation for G(0) G (0). This will furnish the appropriate estimate. Newmans idea is to replace the factor 1/z by (1/z ) + (z/R2 ) in the path integral in (1). Since (G(z ) G (z ))z/R2 is analytic, the value of the path integral along R remains unchanged. We further modify (1) by replacing G(z ) and G (z ) by their respective products with ez . Since ez is entire and has the value 1 at z = 0, we can write

G(0) G (0) =

1 2i

z 1 (G(z ) G (z ))ez ( + 2 ) dz. z R R

Note that for |z | = R we have (1/z ) + (z/R2 ) = (z/|z |2 ) + (z/R2 ) = (2 Re z )/R2 , so that + , (recalling (2) above), if z R 1 z 1 Re z Re z 2 Re z 2 |(G(z ) G (z ))ez ( + 2 )| e e = 2. 2 z R Re z R R Consequently, 1 2i 1 1 z (G(z ) G (z ))ez ( + 2 ) dz z R R

+ R

+ is indeby the M-L theorem. Note that this estimate of the integral along the path R pendent of . Now let us consider the contribution to the integral along R of the integral

Ch: 1

162 (7-12)

TOC

Index

7.3. PROOF OF THE PRIME NUMBER THEOREM

along R . First we use the triangle inequality to obtain the estimate

13

1 2i

1 z (G(z ) G (z ))ez ( + 2 ) dz z R 1 z 1 G(z )ez ( + 2 ) dz + z R 2i 1 z G (z )ez ( + 2 ) dz z R

1 2i

= |I1 (R)| + |I2 (R)|. First consider I2 (R). Since G (z ) is an entire function, we can replace the path of inte gration R by the semicircular path from iR to iR in the left half plane. For z on this semicircular arc, the modulus of the integrand in I2 (R) is

|(

F (t)ezt dt)ez
0

1 2| Re z | 2 2 Re z )| = 2. 2 2 R | Re z | R R

(Note that |F | 1, we can replace the upper limit of integration by , and ex 1 for x 0.) This inequality also holds if Re z = 0 (let z iy ). Thus by the M-L theorem we get |I2 (R)| (1/2 )(2/R2 )(R) = 1/R, again. Finally, we consider |I1 (R)|. This will be the trickiest of all since we only know that on R , G is an analytic extension of the explicitly dened G in the right half plane. To deal with this case, rst choose a constant M (R) > 0 such that |G(z )| M (R) for z R . Choose 1 such that 0 < 1 < (R) and break up the integral dening I1 (R) into two parts, corresponding to Re z < 1 and Re z 1 . The rst contribution is bounded in modulus by 1 1 1 1 1 1 M (R)e1 ( + )R = RM (R)( + )e1 , 2 (R) R 2 (R) R which for xed R and 1 tends to 0 as . On the other hand, the second contribution is bounded in modulus by 1 1 1 1 M (R)( + )2R arcsin , 2 (R) R R the last factor arising from summing the lengths of two short circular arcs on the path of integration. Thus for xed R and (R) we can make the above expression as small as we please by taking 1 suciently close to 0. So at last we are ready to establish the conclusion of this theorem. Let > 0 be given. Take R = 4/ and x (R), 0 < (R) < R, such that G is analytic inside and on R . Then as we saw above, for all , 1 2i and also 1 2i 1 1 z = . (G (z )ez ( + 2 ) dz z R R 4 1 z 1 (G(z ) G (z ))ez ( + 2 ) dz = z R R 4

+ R

Ch: 1

163 (7-13)

TOC

Index

14

CHAPTER 7. THE PRIME NUMBER THEOREM

Now choose 1 such that 0 < 1 < (R) and such that 1 1 1 1 M (R)( + )2R arcsin < . 2 (R) R R 4 Since 1 1 1 RM (R)( + )e1 < 2 (R) R 4 for all suciently large, say 0 , it follows that |G (0) G(0)| < , 0 which completes the proof.

Proof of (7.3.2)
Let f (x) and g (z ) be as in the statement of the corollary. Dene F on [0, +) by F (t) = et f (et ) c. Then F satises the rst part of the hypothesis of the auxiliary Tauberian theorem, so let us consider its Laplace transform, G(z ) =
0

(et f (et ) c)ezt dt,

which via the change of variables x = et becomes G(z ) =

1 1 1

1 dx ( f (x) c)xz x x f (x)xz2 dx c f (x)xz2 dx c z

1

= =

xz1 dx

g (z + 1) c = z+1 z 1 c = [g (z + 1) c]. z+1 z It follows from the hypothesis that g (z + 1) (c/z ) has an analytic extension to a neighborhood of the line Re z = 0, and consequently the same is true of the above function G. Thus the hypotheses of the auxiliary Tauberian theorem are satised, and we conclude that the improper integral 0 F (t) dt exists and converges to G(0). In terms of f , this says that 0 (et f (et ) c) dt exists, or equivalently (via the change of variables x = et once more) that
1

dx f (x) c) x x

Ch: 1

164 (7-14)

TOC

Index

7.3. PROOF OF THE PRIME NUMBER THEOREM

15

exists. Recalling that f is nondecreasing, we can infer that f (x)/x c as x . For let > 0 be given, and suppose that for some x0 > 0, [f (x0 )/x0 ] c 2. It follows that f (x) f (x0 ) x0 (c + 2) x(c + ) for x0 x Hence,
c+2 c+ x0

c + 2 x0 . c+

x0 x

dx f (x) c) ( x x

c+2 c+ x0

x0

c + 2 dx = ln( ). x c+

x) dx But x12 ( f ( x c) x 0 as x1 , x2 , because the integral from 1 to is convergent. Thus for all x0 suciently large,
c+2 c+ x0

(
x0

f (x) dx c + 2 c) < ln( ). x x c+

However, reasoning from the assumption that [f (x0 )/x0 ]c 2, we have just deduced the opposite inequality. We must conclude that for all x0 suciently large, [f (x0 )/x0 ] c < 2. Similarly, [f (x0 )/x0 ] c > 2 for all x0 suciently large. [Say [f (x0 )/x0 ] c 2. The key inequality now becomes f (x) f (x0 ) x0 (c 2) x(c ) for ( c 2 )x0 x x0 c

2 and the limits of integration in the next step are from cc x0 to x0 .] Therefore f (x)/x c as x , completing the proof of both the corollary and the prime number theorem. The prime number theorem has a long and interesting history. We have mentioned just a few of the many historical issues related to the PNT in this chapter. There are several other number theoretic functions related to (x), in addition to the function (x) that was introduced earlier. A nice discussion of some of these issues can be found in Eric W. Weisstein, Prime Number Theorem, from MathWorldA Wolfram Web Resource, http://mathworld.wolfram.com/PrimeNumberTheorem.html. This source also includes a number of references on PNT related matters.

References
1. Karl Sabbach, The Riemann Hypothesis-The Greatest Unsolved Problem in Mathematics, Farrar, Strass and Giroux, New York, 2002. 2. Julian Havil, Gamma-Exploring Eulers Constant, Princeton University Press, Princeton and Oxford, 2003, Chapter 15.

Ch: 1

165 (7-15)

TOC

Index

Solutions
Chapter 1
1. |z1 + z2 |2 + |z1 z2 |2 = (z1 + z2 )(z 1 + z 2 ) + (z1 z2 )(z 1 z 2 ) = 2|z1 |2 + 2|z2 |2 . A diagram similar to Fig. 1.1.1 illustrates the geometric interpretation that the sum of the squares of the lengths of the diagonals of a parallelogram equals twice the sum of the squares of the lengths of the sides. 2. Again, use a diagram similar to Fig. 1.1.1. 3. (a) Let z1 = a + bi, z2 = c + di; then |z1 ||z2 | cos is the dot product of the vectors (a, b) and (c, d), that is, ac + bd = Re z1 z 2 . Also, |z1 ||z2 | sin is the length of the cross product of these vectors, that is, |ad bc| = | Im z1 z 2 |. [Strictly speaking, we should take the cross product of the 3-dimensional vectors (a, b, 0) and (c, d, 0).] (b) The area of the triangle is half the area of the parallelogram determined by z1 and z2 . The area of the parallelogram is the length of the cross product of the vectors (a, b) and (c, d), which is | Im z1 z 2 |.
g y

g 4. Say x exists near (x0 , y0 ) and is continuous at (x0 , y0 ), while (x0 , y0 ). Write, as in (1.4.1),

merely exists at

g (x, y ) g (x0 , y0 ) = g (x, y ) g (x0 , y ) + g (x0 , y ) g (x0 , y0 ). Apply the mean value theorem to the rst dierence and the denition of to the second dierence to obtain g g (x, y )(x x0 ) + (x0 , y0 )(y y0 ) + (y )(y y0 ) x y where x is between x0 and x and (y ) 0 as y y0 . In (1.4.1) we may take A= g (x0 , y0 ), x B= 1 (x, y ) = g g (x, y ) (x0 , y0 ), x x
g y (x0 , y0 )

g (x0 , y0 ), y

2 (x, y ) = (y ).
u x v = 1, y = 1. Thus the Cauchy-

5. We have u(x, y ) = x, v (x, y ) = y , hence Riemann equations are never satised. 1

Ch: 1 2 3 4 5 6 7

166 (Soln-1)

TOC

Index

2 6. Since u(x, y ) = x2 + y 2 , v (x, y ) = 0, the Cauchy-Riemann equations are satised at x = y = 0, but nowhere else. The result follows from (1.4.2) and Problem 4. (Dierentiability at z = 0 can also be veried directly, using the denition of the derivative.)
u 7. Since u(0, y ) = u(x, 0) = 0 for all x, y, u x (0, 0) = y (0, 0) = 0. Take v to be identically 0. If u is real-dierentiable at (0,0), then f = u + iv is complex-dierentiable at (0,0) by (1.4.2). Now dierentiability of f at z0 requires that (f (z ) f (z0 ))/(z z0 ) approach a unique limit as z approaches z0 along an arbitrary path. In the present case, let z 0 along the line y = x. The dierence quotient is 1 if x > 0 x2 i = 1+ 1 x + ix if x < 0. 1+i

Therefore f is not complex-dierentiable at the origin, hence u cannot be realdierentiable there. a b 8. Let Mab = , and let h(a + bi) = Mab . Then h is 1-1 onto and h(z1 + z2 ) = b a h(z1 ) + h(z2 ), h(z1 z2 ) = h(z1 )h(z2 ). The result follows. 9. By (b), either 1 P or 1 P . Since i2 = (i)2 = 1, we have 1 P by (a), hence 1 P by (a) again. But 1 P and 1 P contradicts (b).

11. As in Problem 10, (z ) 0 for all z . Also, x2 = (x2 ) = (x)(x), and consequently (x) = x, x 0. Thus (x) = |x| for real x. If z = x + iy , then (z ) (x) + (i)(y ) = |x| + |y |. (Note that ((i))2 = (i2 ) = (1) = 1, so (i) = 1.) Therefore is bounded on the unit circle, and the result follows from Problem 10. 12. Since |z |2 = (z )(z ) and |1 z |2 = (1 z )(1 z ), we have |z | = |1 z | i zz z z + = 1 z z + zz i zz 1 = (zz 1). Since || < 1, this can happen i zz = 1, that is, |z | = 1.

10. If (z ) < 0, let w2 = z ; by (ii), ((w))2 = (z ) < 0, contradicting (w) R. Thus (z ) 0 for all z . Since (z n ) = [(z )]n by (ii), it follows from (iii) that (z ) 1 for |z | = 1. By (i) and (ii), |z |2 = a(|z |2 ) = (z )(z ), so for z on the unit circle, (z ) < 1 implies (z ) > 1, and therefore (z ) = 1 for |z | = 1. Thus for arbitrary z = 0 we have (z ) = (z/|z |)(|z |) = (|z |) = |z |.

14. To show that u is harmonic, verify directly that 2 u/x2 + 2 u/y 2 = 0. To nd v , use the technique of (1.6.2). In part (i) we have u v = = ey cos x, x y v u = = ey sin x. y x

13. If z = r cos +ir sin , then 1/z = (1/r) cos i(1/r) sin , so z +1/z = (r +1/r) cos + i(r 1/r) sin , which is real i sin = 0 or r 1/r = 0. The result follows.

Thus (using calculus) v (x, y ) = ey sin x. In part (ii) we have v u = = 6xy, x y v u = = 2 3x2 + 3y 2 y x

Ch: 1

167 (Soln-2)

TOC

Index

3 so v (x, y ) = 3x2 y + 2y + y 3 . Note that if z = x + iy , then u + iv can be written as z 3 + 2z . After complex exponentials are studied further in Section 2.3, it will follow that in part (i), u + iv = eiz . 15. (i) Note that |z z0 | = r i |az + b (az0 + b)| = r|a|. (ii) T (0) = 1 + i, so b = 1 + i; r|a| = |a| = 2, so T (z ) = az + 1 + i, |a| = 2. (iii) Since | 2 + 2i| > 2, the desired result cannot be accomplished.

16. Since u = ex , v = 0, the Cauchy-Riemann equations are never satised. 17. We have g (z + h) g (z ) f (z + h) f (z ) = = h h f (z + h) f (z ) h

18. The circle is described by |z z0 |2 = r2 , or, equivalently, (z z0 )(z z 0 ) = r2 ; the result follows. 19. If P (z ) = 0 for some z D(0, 1), then (1 z )P (z ) = 0, that is, (1 z )(a0 + a1 z + + an z n ) = 0, which implies that a0 = (a0 a1 )z + (a1 a2 )z 2 + + (an1 an )z n + an z n+1 . (1)

Thus g is analytic at z i f is analytic at z , and in this case, g (z ) = f (z ). Since z i z , the result follows.

20. If P (z ) = 0 for some z with |z | 1, then |z | = 1 by Problem 19. The only way for (1) in Problem 19 to be satised is if all terms (a0 a1 )z, . . . , (an1 an )z n , an z n+1 are nonnegative multiples of one another (cf. Problem 2), and this requires that z be real, i.e., z = 1. But P (1) = a0 + + an > 0, so there are no roots in D(0, 1).

Since ai ai+1 0, the absolute value of the right side of (1) is at most |z |(a0 a1 + a1 a2 + + an1 an + an ) = a0 |z |. If |z | < 1, this is less than a0 , a contradiction.

Chapter 2
Section 2.1
1. We have (t) = (1 t)(i) + t(1 + 2i) = t + i(3t 1), 0 t 1; thus
1 1

y dz =
0

(Im (t)) (t) dt =

0

(3t 1)(1 + 3i) dt =

1 3 +i . 2 2

2. We have (t) = t + it2 , 1 t 2; thus

2 2

z dz =
1

(t) (t) dt =
1

7 (t it2 )(1 + i2t) dt = 9 + i . 3

Intuitively, z dz =

(x iy )(dx + idy ) =

x dx + y dy + i(x dy y dx).

Ch: 1

168 (Soln-3)

TOC

Index

4 Since y = x2 on , this becomes

2 1

[x dx + x2 (2x dx) + ix(2x dx) ix2 dx]

as above. Note also that, for example,

x dy =

4 2

y dy .

3. The rst segment may be parametrized as (1t)(i)+t(2+5i) = 2t+i(6t1), 0 t 1, and the second segment as (1 t)(2 + 5i) + t5i = 2 2t + 5i, 0 t 1. Thus
1

f (z ) dz =
0 1

[i Im (t) + (Re (t))2 ] (t) dt [i(6t 1) + 4t2 ](2 + 6i) dt

1 0

=
0

[5i + (2 2t)2 ](2) dt

28 8 = + 12i 10i = 12 + 2i. 3 3 4. Since is a path and h is continuously dierentiable, it follows that 1 is a path. We have, with s = h(t),
d d f (1 (t))1 (t) dt =

f (z ) dz =
1 c

f ( (h(t))) (h(t))h (t) dt

c

= f ( (s)) (s) ds =

f (z ) dz.

(Strictly speaking, this argument is to be applied separately to the subintervals on which 1 is continuous.) 5. (a) By (2.1.6), f (z2 ) f (z1 ) =
1 [z1 ,z2 ]

f (z2 ) f (z1 ) = (z2 z1 ) 0 f ((1 t)z1 + tz2 ) dt. Since Re f > 0 by hypothesis, we have Re[(f (z2 ) f (z1 ))/(z2 z1 )] > 0. In particular, f (z1 ) = f (z2 ). (b) For f (z ) = z + 1/z , we have f (z ) = 1 1/z 2 , so in polar form, Re f (rei ) = 1 (cos 2)/r2 , which is greater than 0 i r2 > cos 2. By examining the graph of r2 = cos 2 (a two-leaved rose), w see that for a > 0 and suciently large, and > 0 and suciently small, we have Re f > 0 on = C\A, where A is the set of points inside or on the boundary of the innite triangle determined by the rays [a, (1 )i, ) and [a, (1 )(i), ). Now is starlike and contains i, with f (i) = f (i), which proves that (a) does not generalize to starlike regions. (c) Since f (z0 ) = 0, either Re f (z0 ) = 0 or Im f (z0 ) = 0. If the real part is nonzero, then Re f must be of constant sign (positive or negative) on a suciently small disk centered at z0 . The result then follows from (a). The remaining case is handled by observing that Im f = Re(if ) = Re[(if ) ].

f (w) dw. If w = (1 t)z1 + tz2 , we obtain

Ch: 1

169 (Soln-4)

TOC

Index

Section 2.2
1. The statement about pointwise convergence follows because C is a complete metric space. If fn f uniformly on S , then |fn (z ) fm (z )| |fn (z ) f (z )| + |f (z ) fm (z )|, hence {fn } is uniformly Cauchy. Conversely, if {fn } is uniformly Cauchy, it is pointwise Cauchy and therefore converges pointwise to a limit function f . If |fn (z ) fm (z )| for all n, m N and all z S , let m to show that |fn (z ) f (z )| for n N and all z S . Thus fn f uniformly on S . 2. This is immediate from (2.2.7).
2

3. We have f (x) = (2/x3 )e1/x for x = 0, and f (0) = limh0 (1/h)e1/h = 0. Since 2 f (n) (x) is of the form pn (1/x)e1/x for x = 0, where pn is a polynomial, an induction argument shows that f (n) (0) = 0 for all n. If g is analytic on D(0, r) and g = f on (r, r), then by (2.2.16), g (z ) = n=0 f (n) (0)z n /n!, z D(0, r). [Note that f (n) (0) is determined once f is specied on (r, r).] Thus g , hence f , is 0 on (r, r), a contradiction. 4. (a) The radius of convergence is at least 1/. For if = , this is trivial, and if < , then for a given > 0, eventually |an+1 /an | < + , say for n N . Thus |aN +k z N +k | |aN ||z |N |( + )z |k , k = 0, 1, . . . . By comparison with a geometric series, the radius of convergence is a least 1/( + ). Since is arbitrary, the result follows. Note that the radius of convergence may be greater than 1/. for example, let an = 2 if n is even, and an = 1 if n is odd. The radius of convergence is 1, but lim supn |an+1 /an | = 2, so 1/ = 1/2. (b) The radius of convergence r is exactly 1/. For r 1/ by (a), and on the other hand we have limn |an+1 z n+1 /an z n | = |z |, which is greater than 1 if |z | > 1/. Thus limn an z n cannot be 0, and hence the series cannot converge, for |z | > 1/. [This is just the ratio test; see (2.2.2).] 5. Since an = f (n) (z0 )/n!, we have lim supn |an |1/n lim supn (bn )1/n . The radius of convergence of the Taylor expansion bout z0 is therefore 0, a contradiction. 6. (a) As in (2.2.16), write f (z ) = The term in brackets is z z0 + + w z0 z z0 w z0
n n+1 z z0 wz0 z z0 . 1 w z0

1 2i

f (w) w z0

1 1
z z0 wz0

dw.

1+ By (2.2.11), f (z ) =

n (k ) (z0 )(z k=0 [f

z0 )k /k !] + Rn (z ), where f (w) dw. (w z )(w z0 )n+1

Rn (z ) =

(z z0 )n+1 2i

Ch: 1

170 (Soln-5)

TOC

Index

6 (b) If |z z0 | s < r1 , then |w z | r1 s for all w , hence by (2.1.5), |Rn (z )|

s

|z z0 |n+1 Mf () r1 n+1 2r1 Mf () r s 2 (r1 s)r1 1

s r1

n+1

9. (a) Let an = 1/n, bn (z ) = k=0 z k = (1 z n )/(1 z ) if z = 1. For any xed z with |z | = 1, z = 1, we have |bn (z )| 2/|1 z | < for all n, and the desired result follows from Problem 8(a). n1 ix (b) Let an = 1/n and bn = + ei2x + + ei(n1)x ) = k=0 sin kx = Im(1 + e inx ix Im[(1 e )/(1 e )] (if x is not an integral multiple of 2 ; the series converges to 0 in that case). Now 1 einx 1 eix
2

7. We compute k=r ak bk = ar (br+1 br ) + + as (bs+1 bs ) = ar br + br+1 (ar ar+1 ) + + bs (as1 as ) + as bs+1 as+1 bs+1 + as+1 bs+1 , and the result follows. s s 8. (a) If |bn | M for all n, then k=r |bk+1 ak | M k=r |ak | = M (ar ar+1 + ar+1 ar+2 + + as as+1 ) = M (ar as+1 ) 0 as r, s . The result follows from Problem 7. s (b) By the argument of (a), k=r ak bk (z ) 0 as r, s , uniformly for z S .
n1

1 sin2 (nx/2) 1 cos nx = 2 2 1 cos x sin (x/2) sin (x/2)

which is uniformly bounded on {x : 2k + x (2k + 2) }. The result follows from Problem 8(b). (c) Let sin nz = sin n(x + iy ) = (einz einz )/2i = (einx eny einx eny )/2i. If y = 0, then (1/n) sin nz as n , hence n (1/n) sin nz cannot converge. 10. If z / C+ R, then f (z + h) f (z ) f (z + h) f (z ) = = h h f (z + h) f (z ) h f (z ) as h 0.

Thus f is analytic on C \ R. On R we have z = z and f (z ) = f (z ) = f (z ), so f is continuous on C. 11. The idea is similar to (2.1.12). If T is a triangle in C, express T f (z ) dz as a sum of integrals along polygons whose interiors are entirely contained in C+ or in the open lower half plane C , and at worst have a boundary segment on R. But, for example, f (z ) dz [a,b] f (z ) dz as 0 (use the M-L theorem). It follows that [a+i,b+i ] f (z ) dz = 0, and f is analytic on C by Moreras theorem. T 12. (a) By (2.2.10), F is analytic on C \ C (z0 , r) and F (z ) = C (z0 ,r) (w z )2 dw. But for any xed z , the function h given by h(w) = 1/(w z )2 has a primitive, namely 1/(w z ), on C \ {z }. Thus by (2.1.6), F (z ) = 0. By (2.1.7b), F is constant on D(z0 , r). (b) We have F (z0 ) = 2i [see the end of the proof of (2.2.9)] and thus by (a), F (z ) = 2i for all z D(z0 , r). As in the proof of (2.2.9), 1 2i
C (z0 ,r )

f (w) f (z ) dw = wz 2i

C (z0 ,r )

1 dw = f (z ) by part (b). wz

Ch: 1

171 (Soln-6)

TOC

Index

7 13. (a) This follows from (2.2.11), (2.1.4) and (2.1.2). (b) By part (a) with a = 0, |f (n) (0)/n!| M rk /rn 0 as r , if n > k . Thus the Taylor coecient an is 0 for n > k , and the result follows. (c) The statement and proof of (b) go through even if k is a nonnegative number, not necessarily an integer. Take k = 3/2 to obtain (c). 14. If |z | < 1, then n=0 |an z n | < , hence the radius of convergence r is at least 1. If r > 1, then the series for f (z ), namely n=1 nan z n1 , will converge absolutely when |z | = 1, so n=1 n|an | < , a contradiction. Thus r = 1. . By (2.1.8), 15. Let T be a triangle such that T f (z ) dz = 0 for every n. Since T n fn f uniformly on T , we have T f (z ) dz = 0. By Moreras theorem, f is analytic on .

Section 2.3
1. If u + iv = sin(x + iy ), then u = sin x cosh y and v = cos x sinh y . If y = b, then (u2 / cosh2 b) + (v 2 / sinh2 b) = 1. Thus {x + iy : /2 < x < /2, y > 0} is mapped onto {u + iv : v > 0}, {x + iy : /2 < x < /2, y < 0} is mapped onto {u + iv : v < 0}, {x + iy : x = /2, y 0} is mapped onto {u + iv : v = 0, u 1}, and nally {x + iy : x = /2, y 0} is mapped onto {u + iv : v = 0, u 1}, and the mapping is one-to-one in each case. Since sin(z + ) = sin z , the statement of the problem follows. 2. If sin(x + iy ) = 3, then sin x cosh y = 3, cos x sinh y = 0. If sinh y = 0 then y = 0, cosh y = 1, sin x = 3, which is impossible. Thus cos x = 0, x = (2n + 1)/2. If n is odd then sin x = 1, cosh y = 3, again impossible. Thus the solutions are z = x + iy where x = (4m + 1)/2, m an integer, y such that cosh y = 3 (two possibilities, one the negative of the other). 3. Since sin z = z z 3 /3! + x5 /5! , the only nonzero contribution to the integral is the single term C (0,1) dz/3!z = 2i/6 = i/3.

5. If f (z ) = n=0 an z n , then since f + f = 0 we have n(n 1)an + an2 = 0, n = 2, 3 . . . . Since f (0) = 0 and f (0) = 1, we have a0 = 0, a1 = 1, hence a2 = 0, a3 = 1/3!, a4 = 0, a5 = 1/5!, and so on. Thus f (z ) = z z 3 /3! + z 5 /5! z 7 /7! + = sin z . 6. As in Problem 5, nan an1 = 0, a0 = 1. Therefore f (z ) = 1 + z + z 2 /2! + = ez .

4. This follows from two observations: (a) 1 + z + z 2 /2! + + z n /n! ez as n , uniformly for |z | r; (b) min|z|r |ez | > 0.

Section 2.4
2. If f (z ) = (z z0 )m g (z ) on with g (z0 ) = 0, expand g in a Taylor series about z0 to conclude that aj = 0 for j < m and am = 0. Conversely, if a0 = = am1 = 0, am = 0, then f (z ) = n=m an (z z0 )n = (z z0 )m g (z ) with g (z0 ) = 0. (Strictly speaking, this holds only on some disk D(z0 , r), but g may be extended to all of by the formula f (z )/(z z0 )m .) The remaining statement of (2.4.5) follows from (2.2.16). 1. Take f (z ) = sin(1/z ), = C \ {0}; then f has zeros at 1/n 0 / .

Ch: 1

172 (Soln-7)

TOC

Index

8 3. Let f be continuous on the region . (i) If f satises (b) of the maximum principle, f need not satisfy (a). For example, let = D(0, 2) and f (z ) = 1, |z | 1; f (z ) = |z |, 1 < |z | < 2. (ii) If f satises (c), then f satises (b), hence (b) and (c) are equivalent (assuming is bounded). This is because f must satisfy (c) if we take M = , and consequently f satises (b). (iii) If is bounded and f is continuous on , then (d) implies (b), hence in this case (b), (c) and (d) are equivalent. For let z0 be a point on the boundary of such that |f (z0 )| = M0 = max{|f (z )| : x }. Since z0 , there is a sequence of points zn with zn z0 , hence |f (zn )| |f (z0 )| = M0 . Thus = sup{|f (z )| : z } M0 . If |f | < M0 on , then |f | < on . am (z z0 )m + am+1 (z z0 )m+1 + f (z ) = g (z ) bn (z z0 )n + bn+1 (z z0 )n+1 + where am = 0, bn = 0 (that is, f has a zero of order m Then am /bm f (z ) f (z ) lim = lim = 0 z z0 g (z ) z z0 g (z ) and g a zero of order n at z0 ). if m = n if m > n . if m < n

4. In a neighborhood of z0 , we have

(To handle the last case, apply the second case to g/f .) 5. Immediate from (2.4.12). 6. Im f = 0 on D, hence by part (d) of the maximum and minimum principles for harmonic functions (see (2.4.15) and its accompanying remark), Im f (z ) = 0 for all z D. Thus f is constant on D by the Cauchy-Riemann equations. 7. By the maximum principle, we need only consider K . Now sin(x+iy ) = sin x cosh y + i cos x sinh y . If x = 0 or 2 , then sin(x + iy ) = i sinh y . If y = 0, then sin(x + iy ) = sin x. If y = 2 , then sin(x + iy ) = cosh 2 sin x + i sinh 2 cos x. Since cosh 2 > sinh 2 > 1, it follows that the maximum modulus is attained at x = /2 or 3/2, y = 2 , and max |f | = cosh 2 .

8. Choose z0 K such that |f (z0 )| = max{|f (z )| : z K }. If z0 K , we are nished, so assume that z0 K 0 . By (2.4.12a), f is constant on the component 0 of that contains z0 , which proves the furthermore part. To see that |f (z0 )| = max{|f (z )| : z K }, note that by continuity, f must also be constant on 0 K 0 K . Since 0 is bounded, its boundary is not empty. Choose any z1 0 . Then f (z0 ) = f (z1 ) since z1 0 , so |f (z1 )| = max{|f (z )| : z K }. But z1 K and z1 is not an interior point of K . (If z1 K 0 then D(z1 , r) K 0 for some r > 0, and it would not be possible for z1 to be a boundary point of a component of K 0 .) Consequently, z1 K , and max{|f (z )| : z K } = |f (z1 )| max{|f (z )| : z K } max{|f (z )| : z K }. The result follows.

Ch: 1

173 (Soln-8)

TOC

Index

9 9. By Problem 8, maxz |f (z )| = maxz |f (z )|. But = \ = , and the result follows. 10. Take u = Im f where f is a nonconstant entire function that is real-valued on R. For example, f (z ) = ez , u(x + iy ) = ex sin y ; or f (z ) = z, u(x + iy ) = y . 11. If is disconnected and A is a component of , let f (z ) = 1 if z A, and f (z ) = 0 if z / A. Let g (z ) = 0 if z A, and g (z ) = 1 if z / A. Then f g 0 but f 0, g 0. Assume connected, and let f, g be analytic on with f g 0. If f (z0 ) = 0, then f is nonzero on some disk D(z0 , r), hence g 0 on D(z0 , r). By (2.4.8), g 0 on .

12. The given function can be extended to a function f analytic on S {z : Im z < 0} by the technique of the Schwarz reection principle (2.2.15). Since f (z ) = z 4 2z 2 for z (0, 1), the identity theorem (2.4.8) implies that this relation holds for all z S {z : Im z < 0}. Thus if z S we have f (z ) = z 4 2z 2 , and in particular, f (i) = 3. 13. Apply Liouvilles theorem to 1/f . 14. No, by the identity theorem. If S is an uncountable set, then innitely many points of S must lie in some disk D(0, r), hence S has a limit point. 15. Fix the real number . Then sin( + ) sin cos cos sin is an analytic function of , and is zero for real , hence is identically zero by the identity theorem. A repetition of this argument with xed and variable completes the demonstration. 16. If f = u + iv , then |eif | = ev 1 (because v 0 by hypothesis). By Liouvilles theorem, eif is constant, hence |eif | = ev is constant, so v is constant. But then by the Cauchy-Riemann equations, u is constant, so f is constant. 17. We have (f /g ) = (gf f g )/g 2 , and by the identity theorem, gf f g is identically zero on D(0, 1). The result follows. 18. By Liouvilles theorem, f (z ) ez sin z = c where |c| < 4. Since f (0) = 0 we have c = 0, so f (z ) = ez sin z . 19. By the maximum and minimum principles for harmonic functions, Re(f g ) is identically zero. Therefore f g is constant. 20. If f is never 0 and {f (zn )} is unbounded whenever |zn | 1, then 1/f (z ) 0 as |z | 1. By the maximum principle, 1/f 0, a contradiction.

22. If f is never 0 in D(0, 1) then by the maximum principle, max|z|1 |1/f (z )| < 1, hence |1/f (0)| < 1. This contradicts f (0) = i.

21. Let f = u + iv with f analytic on C. Then |ef | = eu e0 = 1, hence |ef | 1. By Liouvilles theorem, ef is constant. But then |ef |, hence |ef |, is constant. Since |ef | = eu , the result follows.

23. If z = x + iy then u = Re z 3 = x3 3xy 2 . By the maximum principle, it suces to consider u on each of the four line segments forming the boundary of the square. By elementary calculus we nd that the maximum value is 1 and occurs at x = 1, y = 0.

24. If K is a compact subset of D(0, 1), then K D(0, r) for some r (0, 1). If M = max{|f (z )| : |z | r}, then [|f (rz )|/M ] 1 on D(0, 1). By (2.4.16), [|f (rz )|/M ] |z | on D(0, 1). Make the substitution z = w/r to obtain |f (w)| M |w|/r on D(0, r),

Ch: 1

174 (Soln-9)

TOC

Index

10 hence on K . The result now follows from the uniform convergence of the series |w|n , |w| r.

25. If zn z C (0, 1), then for some k , eik z A1 , so f (eik zn ) 0, and therefore F (zn ) 0. By (2.4.12c), F 0. Now for any z D(0, 1), F (z ) = 0, so f (eih z ) = 0 for some h = 0, 1, . . . , n. Thus f has uncountably many zeros, hence a limit point of zeros, in D(0, 1). By the identity theorem, f 0.

26. (a) By Problem 9, {fn } is uniformly Cauchy on , hence by (2.2.4), {fn } converges uniformly on . By (2.2.17), f is analytic on (and continuous on by the uniform convergence). The proof of (2.2.17), in particular the formula (2.2.11), may be (k ) adapted to show that each derivative fn extends to a continuous function on , and (k ) that fn f (k) uniformly on for all k . (b) If p1 , p2 , . . . are polynomials and pn f uniformly on C (0, 1), then by (a), pn converges uniformly on D(0, 1) to a limit function g , where g is analytic on D(0, 1) and continuous on D(0, 1) (and of course g = f on C (0, 1). Conversely, if f is the restriction to C (0, 1) of such a function g , then f can be uniformly approximated by polynomials. To see this, let {rn } be an increasing sequence of positive reals converging to 1, and consider gn (z ) = g (rn z ), |z | < 1/rn . Since gn is analytic on D(0, 1/rn ), there is a (Taylor) polynomial pn such that |pn (z ) gn (z )| < 1/n for |z | 1. But gn converges uniformly to g on D(0, 1) by uniform continuity of g on D(0, 1). The result follows.

Chapter 3
Sections 3.1 and 3.2
1. (a) This follows because log is discontinuous on the ray R [see (3.1.2b)]. (b) Let U be as indicated in Figure S3.2.1, and dene g (z ) = ln |z | + i(z ), 0 < 2, for z 1 ln |z | + i(z ), < 3, for z 2 .

Locally, g (z ) coincides with one of the elementary branches of log z , hence g is an analytic version of log z on . 2. First, we show that f does not have an analytic logarithm on . For f (z )/f (z ) = [1/(z a)] + [1/(z b)], so that if describes a circle enclosing both a and b, (3.2.3) yields [f (z )/f (z )] dz = 2i(n(, a) + n(, b)) = 4i = 0. By (3.1.9), f does not have an analytic logarithm on . However, f has an analytic square root. For if 0 is the angle of [a, b] (see Figure S3.2.2), then dene 1 (z a)1/2 = |z a|1/2 exp(i arg(z a)) 2 1 (z b)1/2 = |z b|1/2 exp(i arg(z b)) 2

Ch: 1

175 (Soln-10)

TOC

Index

11

Figure S3.2.1

S
b a
Figure S3.2.2

Ch: 1

176 (Soln-11)

TOC

Index

12 where the angles are chosen in the interval [0 , 0 +2 ). Then g (z ) = (z a)1/2 (z b)1/2 is the desired analytic square root. The key intuitive point is that if z traverses a circle enclosing both a and b, then the arguments of (z a)1/2 and (z b)1/2 each change by , so that g (z ) returns to its initial value. This shows that the set S = {z : z a = rei0 , r > |b a|} is not a barrier to analyticity.

Remark
f (z ) = z 2 on D(0, 1) \ {0} gives an easier example of an analytic function that is never 0 and has an analytic square root, but not an analytic logarithm. 3. By (3.1.11) we have (a) implies (b), and (b) implies (c) is obvious. To prove that (c) k implies (a), let gk be analytic on with gk = f . Then f /f = kgk /gk , so that if is a closed path in , we have 1 2i
1 gk (z ) dz = gk (z ) 2ik

f (z ) dz 0 f (z )

as k through an appropriate subsequence. By (3.2.3), n(gk , 0) 0 as k . (z ) Since the index must be an integer, n(gk , 0) = 0 for large k . Therefore f f (z ) dz = 0, and the result follows from (3.1.9).
4. As in (3.2.4d), 0 / 1 . If = 2 /1 , then |1 | < 1 + | |, which implies that 2 (t) can never be real and negative. Thus Arg is a continuous argument of , hence n(, 0) = 0. As in (3.2.4d), n(1 , 0) = n(2 , 0).

The hypothesis is satised by all possible values of 1 (t) and 2 (t) except those lying on a line through the origin, with 1 (t) and 2 (t) on opposite sides of the origin. Thus if initially the angle between 1 (t) and 2 (t) (visualizing a complex number z as a vector in the plane pointing from 0 to z ) is less than and remains less than for all t, then 1 and 2 have the same net number of revolutions about 0. The interpretation of the hypothesis is that the length of the leash is always less than the sum of the distances of man and dog from the tree. 5. Suppose is a continuous argument of f . Let (t) = eit , 0 t 2 . Since z = |z |ei(z) = ei(z) when |z | = 1, we have eit = ei(e ) , 0 t 2. Thus t and (eit ) are each continuous arguments of , so by (3.1.6c), (eit ) = t + 2k for some integer k . Let t 0 to obtain (1) = 2k , and let t 2 to obtain (1) = 2 + 2k , a contradiction. Note that (z ), z S , if it is to exist, must be a continuous function of z , that is, a continuous function of position in the plane, as opposed to (eit ), 0 t 1, which is a continuous function of the time parameter t. If we specify (1) = 0 and move around the circle, continuity requires that (1) = 2 , which produces a contradiction. (A function is a function is a function.)
it

Ch: 1

177 (Soln-12)

TOC

Index

13 6. Since f is (uniformly) continuous on S , m = min{|f (z )| : z S } > 0, and for some > 0 we have |f (z ) f (z )| < m whenever z, z S, |z z | < . Break up S into closed squares whose diagonal is less than . Let A be a particular square, and pick any z0 A. Then |z z0 | < for every z A, hence |f (z ) f (z0 )| < m. Thus f (z ) D(f (z0 ), m), and it follows just as in (3.1.7) that f has a continuous logarithm on A. Now let A and B be adjacent squares having a common side. If eg1 = f on A and eg2 = f on B , then for some integer k , g1 g2 = 2ik on the common side. If we replace g2 by g2 + 2ik , we produce a continuous logarithm of f on A B . Continuing in this fashion, we may construct a continuous logarithm on each horizontal strip of S , and then piece the horizontal strips together to cover all of S . Formal details are not dicult to supply. Remark: The same technique works if S is an innite rectangular strip. 7. If D is a disk contained in , then f has an analytic logarithm h on D. Thus g h is constant on D, so g is analytic on D, hence on all of . 8. We will show that f and g are entire functions such that f 2 + g 2 = 1 i for some entire function h, we have f = cos h and g = sin h. The if part is immediate, so consider the only if assertion. Since f + ig is entire and never 0, f + ig has an analytic logarithm h0 on C. If h = ih0 , then f + ig = eih and f ig = (f + ig )1 = eih . Consequently, f = (eih + eih )/2 = cos h and g = (eih eih )/2i = sin h. 9. (a) Since (f /g )n = 1, f /g is a continuous map of S into {ei2k/n : k = 0, 1, . . . , n 1}. Since S is connected, the image must be connected. Therefore the image consists of a single point, so f /g = ei2k/n for some xed k. (b) Take S = [1, 1] and let f (x) = g (x) = x, 0 x 1; f (x) = i |x|, 1 x 0; g (x) = i |x|, 1 x 0. Then f 2 (x) = g 2 (x) = x for all x S .

Section 3.3
1. By (i) of (3.3.1),
)f (z ) ] dw = 0, and the result follows from (3.2.3). [ f (w wz

2. In order to reproduce the proof in the text, two key observations must be made. (a) Theorem 3.2.3 holds when is a cycle [this was noted in (3.3.5)]. (b) For any cycle = k1 1 + + km m , we have n(, z ) = 0 for all suciently large |z |. This holds because if |z | is large enough, then for each j , z will be in the unbounded . Thus n(, z ) = 0 by (3.2.5). component of C \ j With these modications, the proof in the text goes through. 3. By (3.2.5), n(, z ) is locally constant, and the result then follows from (2.2.10) and (ii) of (3.3.1). 4. By partial fraction expansion, 1 1 1/2 1/2 = = . z2 1 (z 1)(z + 1) z1 z+1

Ch: 1

178 (Soln-13)

TOC

Index

14 By the Cauchy integral formula [(ii) of (3.3.1)], 1 1 1 dz = 2i( ) = 0. z2 1 2 2

5. Apply Problem 3 with k = 3, f (w) = ew + cos w, z = 0, to obtain 2i i ez + cos z dz = n(j , 0)f (3) (0) = n(j , 0). z4 3! 3

6. By (3.3.7), we may replace by 0 (t) = cos t + i sin t, hence But dz = z

2 0

Since n(1 , 0) = 1, n(2 , 0) = 2, the integral on 1 is i/3 and the integral on 2 is 2i/3.

dz/z =

dz/z = 2i.

a sin t + ib cos t dt. a cos t + ib sin t

Take imaginary parts to obtain

2

2 = ab
0

cos2 t + sin2 t dt, a2 cos2 t + b2 sin2 t

and the result follows.

Section 3.4
\ = {0, }, which is not connected. If f (z ) = 1/z on 1. (a) Let = C \ {0}. Then C and describes any circle with center at 0, then f (z ) dz = 2i = 0. (b) Let be the union of two disjoint disks D1 and D2 . Then is disconnected, but \ is connected. C

2. No. For example, the situation illustrated in Figure 3.4.4 can occur even if is connected. In this case, there is no way to replace the cycle by a single closed path.

3. (a) Since 1 z is analytic and never 0 on the simply connected open set C \ 1 , it has an analytic square root f . If we specify that 1 = 1, then f is determined uniquely, by Problem 9(a) of Section 3.2. A similar analysis applies to g . (b) Since f 2 = g 2 and f (0) = g (0) = 1, f = g on any connected open set containing 0, by Problem 9(a) of Section 3.2. In particular, f = g below . Suppose f = g above . Since f is analytic on 2 \ {1} and g is analytic on 1 \ {1}, f can be extended to a function analytic on C \ {1}. Thus 1 z has an analytic square root on C \ {1}, so that z has an analytic square root on C \ {0}, a contradiction. (If h2 (z ) = z with h continuous on C \ {0}, then h(eit ) = eit/2 k (t), where k (t) = 1. A connectedness argument shows that either k (t) 1 or k (t) 1, and in either case we obtain a contradiction by letting t 0 and t 2 .) It follows that f (z0 ) = g (z0 ) for at least one point z0 above , and as above, we must have f = g at all points above . (c) The function h may be obtained by expanding g in a Taylor series on D(0, 1). Thus h(z ) = f (z ), z D(0, 1), z below 1 , and h(z ) = f (z ), z D(0, 1), z above 1 .

Ch: 1

179 (Soln-14)

TOC

Index

15 . 4. It follows from part (a) of (3.4.3) that if C, then is open in C i is open in C \ K is open in C . On the other hand, if V , If K is a compact subset of C, then C , then C \ V is a closed and bounded (hence compact) subset of C. where V is open in C are of two types: (i) open subsets of C, and (ii) complements Thus the open sets in C (with the topology induced by the chordal of compact subsets of C. Consequently, C metric) is homeomorphic (via the identity map) to the one point compactication of C.

Chapter 4
Section 4.1
1. (a) If f has a removable singularity at z0 , then as in the proof of (4.1.5), f can be dened or redened at z0 so as to be analytic on D(z0 , r) for some r > 0. Thus f is bounded on D (z0 , ) for some > 0, in fact f is bounded on D(z0 , ). Conversely, if f is bounded on D (z0 , ), let g (z ) = (z z0 )f (z ). Then g (z ) 0 as z z0 , so by the rst equivalence of (4.1.5a), g has a removable singularity at z0 . Since the Laurent expansion of g has only nonnegative powers of z z0 , it follows that f has either a removable singularity or a pole of order 1 at z = z0 . The second case is impossible by the rst equivalence of (4.1.5b), and the result follows. (b) If f has a pole of order m at z0 , then (z z0 )m f (z ) K = 0 as z z0 , so |f (z )| . Conversely, if |f (z )| as z z0 , then by (4.1.5a) and (4.1.6) (which we use instead of (4.1.5c) to avoid circularity), f cannot have a removable or essential singularity at z0 , so f must have a pole. 2. (a) Since limzn (z n )z/ sin z = n/ cos n = (1)n n , there are, by (4.1.5), simple poles at z = n , n a nonzero integer. Since z/ sin z 1 as z 0, there is a removable singularity at z = 0. Now f (1/z ) = 1/z sin 1/z has poles at z = 1/n, n = 1, 2, . . . , so 0 is a nonisolated singularity of f (1/z ), hence is a nonisolated singularity of f (z ). (b) There is an isolated essential singularity at 0 since e1/x as x 0+ , e1/x 0 as x 0 . There is a removable singularity at since ez is analytic at 0. (c) There is an isolated essential singularity at 0 since z cos 1/z = z (1 1/2!z 2 + 1/4!z 4 ), z = 0. There is a simple pole at because (1/z ) cos z has a simple pole at 0. (d) There is a pole of order 2 at 0 since z 2 f (z ) 1 as z 0. There are poles of order 1 at z = i2n, n = 1, 2, . . . since (z i2n )/z (ez 1) (1/i2n )(1/ei2n ) = 1/i2n as z i2n . (e) There are simple poles at z = n, n = 0, 1, 2, . . . since (z n ) cos z/ sin z cos n/ cos n = 1 as z n . There is a non-isolated singularity at because is a limit point of poles. 3. We have f (z ) =
1 z

3 z +1

2 z 2 ,

and

1 1 1 = = = (z + 1)n , z (z + 1) 1 1 (z + 1) n=0

|z + 1| < 1;

Ch: 1

180 (Soln-15)

TOC

Index

16 2 2 2 2/3 = (1/3)n (z + 1)n , = = z2 (z + 1) 3 3 n=0 ( z + 1) 1 1 3 Thus f (z ) = 3 2 1 + n+1 (z + 1)n , z + 1 n=0 3

|z + 1| < 3.

0 < |z + 1| < 1.

We may obtain a Laurent expansion for 1 < |z + 1| < 3 by modifying the expansion of 1/z , as follows: 1 1 1 1 = = z+11 = , z (z + 1) 1 z + 1 n=0 (z + 1)n 1 z+1 Therefore f (z ) = 2 + z+1
1

|z + 1| > 1.

k=1

1 2 (1/3)n (z + 1)n , k (z + 1) 3 n=0

n=0

1 < |z + 1| < 3.

For |z +1| > 3, the expansion 1/z = of 2/(z 2) must be modied:

1/(z +1)n+1 is acceptable, but the expansion

2 2 2 3n (z + 1)n . = = z+13 = z2 (z + 1) 3 z + 1 n=0 1 z+1 Thus f (z ) = 4. We have 1 1/2 (1)n z n /2n+1 , = = z+2 1 + z/2 n=0 and 1 1/z (1)n 2n /z n+1 , = = z+2 1 + 2/z n=0

3 1 + 2(3n1) 1 3n1 + +2 = , n n z + 1 n=1 (z + 1) (z + 1) (z + 1)n n=1 n=2

|z + 1| > 3.

|z | < 2

|z | > 2.

Now 1/(1 z ) = n=0 z n , |z | < 1, and therefore by dierentiation, 1/(1 z )2 = n1 , |z | < 1. Also n=1 nz 1 1/z 1 = = , n+1 1z 1 1/z z n=0

|z | > 1,

Ch: 1

181 (Soln-16)

TOC

Index

17 hence 1 n+1 = , (1 z )2 z n+2 n=0 Thus f (z ) = = = 1 [n + 1 + (1)n 2(n+1) ]z n , + z n=0

|z | > 1.

0 < |z | < 1 1 < |z | < 2

1 n+1 (1)n 2(n+1) z n , + + z n=0 z n+2 n=0 2 1 [n 1 + (1)n1 2n1 ] n , + z n=2 z

|z | > 2.

Remark
The coecients of the Taylor expansion of f (1/z ) about z = 0 are the same as the coecients of the Laurent expansion of f (z ) valid for |z | > 2, that is, in a neighborhood of . For this reason, the expansion of f (z ) for |z | > 2 may be called the Taylor expansion of f about .

5. Since g (z ) = z/(ez ez ) 1/2 as z 0, g has a removable singularity at z = 0. We may compute the derivatives of g at z = 0 to form the Taylor expansion g (z ) = (1/2) (1/12)z 2 + (7/720)z 4 , 0 < |z | < . Thus f (z ) = Alternatively, g (z ) = 1 2 1+ z4 z2 + + 3! 5!
1

z 2 (ez

1 1 7 1 = 3 + z + , 0 < |z | < . z e ) 2z 12z 720

=
n=0

an z n ,

and the Taylor coecients may be found by ordinary long division, or by matching coecients in the equation 1+ 6. The function 1 1 1 1 + + sin z z z z+ is analytic for |z | < 2 , hence has a Taylor expansion 1 1 1 z z z+ z4 z2 + + 3! 5! (a0 + a1 z + a2 z 2 + ) = 1 . 2

n=0

an z n . Also,

Ch: 1

182 (Soln-17)

TOC

Index

18 has a Laurent expansion n= bn z n , < |z | < 2 ; the expansion may be found by the procedure illustrated in Problems 3 and 4. Addition of these two series gives the Laurent expansion of 1/ sin z, < |z | < . R(z ) = A1,n1 1 A1 , 1 A1,0 + + + R1 (z ), + (z z1 )n1 (z z1 )n1 1 z z1

7. Expand R(z ) in a Laurent series about z = z1 :

where R1 is analytic at z1 and the representation is valid in some deleted neighborhood of z1 . Dene n1 1

R1 (z ) = R(z )

i=0

A1,i . (z z1 )n1 i

Then R1 is a rational function whose poles are at z2 , . . . , zk with orders n2 , . . . , nk , near z1 . Similarly, expand and R1 has a removable singularity at z1 since R1 = R1 R1 (z ) in a Laurent series about z2 to obtain
n2 1

R2 (z ) = R1 (z )

i=0

A2,i , (z z2 )n2 i

where R2 is a rational function with poles at z3 , . . . , zk with orders n3 , . . . , nk . Continue in this fashion until we reach Rk :
nk 1

Rk (z ) = Rk1 (z )

i=0

Ak,i Bi (z ). = R(z ) (z zk )nk i i=1

Now Rk1 has a pole only at zk , so Rk is a rational function with no poles, that is, a polynomial. But R(z ) 0 as z by hypothesis (deg P < deg Q), and Bi (z ) 0 as z by construction. Thus Rk (z ) 0. Finally,
z zj

lim

dr [(z zj )nj Bm (z )] = 0, dz r

m = j,

and when m = j , the limit is lim dr dz r

nj 1

z zj

i=0

Aj,i (z zj )i = r!Aj,r .

Hence r!Aj,r = lim as desired. Now A C D B 1 = + + + , where z (z + i)3 z (z + i)3 (z + i)2 z+i
z zj

dr [(z zj )nj R(z )] dz r

Ch: 1

183 (Soln-18)

TOC

Index

19 A = [zR(z )]z0 = 1 =i i3 1 =i i 1 z2 1 z3 =1
z i

B = (z + i)3 R(z )]zi = d [(z + i)3 R(z )] dz

C=

=
z i

D= Thus

1 d2 [(z + i)3 R(z )] 2! dz 2

=
z i

z i

= i.

i 1 i i 1 = + + . 3 3 2 z (z + i) z (z + i) (z + i) z+i 8. The series converges absolutely on U = {x + iy : 1 < y < 1}, uniformly on {x + iy : 1 + y 1 } for every > 0, hence uniformly on compact subsets of U . The series diverges for z / U . For

en einz =
n=0 n=0

e(iz1)n =

1 1 eiz1

if |eiz1 | < 1, that is, e(y+1) < 1, or y > 1; this series diverges if |eiz1 | 1, that is, y 1. The convergence is uniform for |eiz1 | r < 1, that is, y 1 . Similarly,

en einz =
n=0

1 1 eiz1

if |eiz1 | < 1, that is, y < 1, with uniform convergence for y 1 . The result follows; explicitly, we have

en sin nz =
n=0

1 1 1 , 2i 1 eiz1 1 eiz1

z U.

is compact, f is bounded. The result follows from Liouvilles theorem. 9. (a) Since C (b) If f (z ) = m=0 bm z m , z C, then g (z ) = f (1/z ) = m=0 bm z m , z C, z = 0, a Laurent expansion of g about z = 0. By (4.1.3), bm = hence |bm | max{|g (z )| : |z | = 1/r}(1/r)m = max{|f (z )| : |z | = r}(1/r)m , 1 2i g (w)wm1 dw,
|z |=1/r

Ch: 1

184 (Soln-19)

TOC

Index

20 which approaches 0 as r if m > k . Thus bm = 0 for m > k , and the result follows. (c) The argument is the same as in (b), except now we know by hypothesis that bm is nonzero for only nitely many m, so it is not necessary to use (4.1.3). (d) Let the poles of f in C be at z1 , . . . , zk , with orders n1 , . . . , nk . (If f had innitely . Let g (z ) = many poles, there would be a nonisolated singularity somewhere in C k f (z ) j =1 (z zj )nj . Then g is analytic on C and has a nonessential singularity at . By (c), g is a polynomial, hence f is a rational function. 10. (a) Pole of order 2 at z = 0, isolated essential singularity at . (b) Isolated essential singularity at z = 0, pole of order 1 at z = 1, removable singularity at . (c) z csc z 1 as z 0, hence csc z k/z has poles of order 1 at z = n, n = 1, 2, . . . , and a pole of order 1 at z = 0 as long as k = 1. If k = 1, there is a removable singularity at z = 0. The point at is a nonisolated singularity. (d) If z is real and near 2/[(2n + 1) ], n = 0, 1, 2, . . . , then exp[sin(1/z )(cos(1/z )] will be near or 0 depending on the sign of z [2/(2n +1) ]. By (4.1.5), exp[tan(1/z )] has an isolated essential singularity at z = [2/(2n + 1) ]. There is a nonisolated singularity at 0 and a removable singularity at . (e) sin(x + iy ) = n when sin x cosh y + i cos x sinh y = n + i0. Thus if n = 1, 2, . . . , then y = cosh1 n, x = (4k + 1)/2, k an integer. (cosh1 n refers to the two numbers u and u such that cosh u = n .) If n = 1, 2, . . . , then y = cosh1 (n ), x = (4k + 3)/2, k an integer. If n = 0, then x = k, y = 0, k an integer. If z0 = x0 + iy0 is any of these points, then by Problem 4 of Section 2.4,
z z0

lim

1 1 z z0 = = . sin(sin z ) cos(sin z0 ) cos z0 cos n cos z0

Now cos(x0 + iy0 ) = cos x0 cosh y0 i sin x0 sinh y0 , and this is nonzero, by the above argument. Thus all the points are poles of order 1. The point at is a nonisolated singularity. 11. Let f (z ) = (z a)/(z b) and U = C \ [a, b]. For any closed path in U , 1 2i

f (z ) 1 dz = f (z ) 2i

1 1 za zb

dz = n(, a) n(, b) = 0

because a and b lie in the same component of C \ . (Note that U , hence [a, b] = .) By (3.1.9), f has an analytic logarithm g on U . Now g = f /f [see (3.1.9)], hence g (z ) = 1 1 za zb (an bn ) = , z n+1 n=0

|z | > max(|a|., |b|) zn, |z | < min(|a|, |b|).

=
n=0

1 1 n+1 bn+1 a

Ch: 1

185 (Soln-20)

TOC

Index

21 Thus g (z ) = log and g (z ) = k + 1 n n=1

za zb

=k+

bn an , nz n n=1

|z | > max(|a|, |b|), z U

1 1 n n b a

|z | < min(|a|, |b|), z U

where k is any logarithm of 1 and k is any logarithm of a/b. 12. If f (C) is not dense in C, then there is a disk D(z0 , r) such that D(z0 , r) f (C) = . Thus for all z C, |f (z ) z0 | r, and the result now follows from Liouvilles theorem applied to 1/[f (z ) z0 ].
n

13. If P (z ) = j =0 aj z j , then P (f (z )) = an [f (z )]n + + a1 f (z ) + a0 . By hypothesis, (z )m f (z ) approaches a nite nonzero limit as z , hence so does (z )mn [f (z )]n . But if j < n, then (z )mn f (z )j = (z )mn f (z )n /[f (z )]nj 0 as z ; the result follows.

Section 4.2
1. By (4.2.7), n(f , 0) = 1. Geometrically, as z traverses once in the positive sense, the argument of z 1 changes by 2 , the argument of z + 2i also changes by 2 , and the argument of z 3 + 4i has a net change of 0. Thus the total change in the argument of f (z ) is 2 2(2 ) = 2 , hence n(f , 0) = 1. 2. Let describe the contour of Figure S4.2.1, with r very large. Now f (z ) = z 3 z 2 + 3z + 5, so f (iy ) = 5 + y 2 + i(3y y 3 ). Thus f is as indicated in Figure S4.2.2. Note that in moving from B to C , the argument of z changes by . Since f (z ) = z 3 (1 z 1 + 3z 2 + 5z 3 ) = z 3 g (z ) where g (z ) 1 as z , the argument of f (z ) changes by approximately 3 . Note also that f (z ) = f (z ), so that f is symmetrical about the real axis. It follows that n(f , 0) = 2, so that f has two roots in the right half plane. (In fact f (z ) = (z + 1)[(z 1)2 + 4], with roots at 1, 1 + 2i, 1 2i.)

y C r x B
Figure S4.2.1

Ch: 1

186 (Soln-21)

TOC

Index

22

B'

A'

C'

Figure S4.2.2 3. Let f (z ) = an z n + + a1 z + a0 , an = 0, g (z ) = an z n . Then if describes a suciently large circle centered at the origin, |f g | < |g | on , so by Rouch es theorem, f has exactly n zeros inside , counting multiplicity. 4. (a) Integrate f (z ) = zeiaz /(z 4 + 4) around the contour indicated in Figure S4.2.3. Then f (z ) dz = 2i residues of f at poles in the upper half plane. The poles of i/4 i3/4 i5/4 i7/4 , 2e , 2e , 2e . The residue at z = z0 is f are at 2e lim (z z0 )zeiaz z0 eiaz0 = 3 . 4 z +4 4z0

z z0

Thus 2i exp(ia 2ei/4 ) exp(ia 2ei3/4 ) + f (z ) dz = 4 2ei/2 2ei3/2 which reduces to [exp(ia 2( 2/2 + i 2/2)) exp(ia 2( 2/2 + i 2/2))] 4 = a ia e (e eia ) = iea sin a. 4 2

An application of (2.1.5) shows that the integral of f around the semicircle approaches 0 as r . Thus in the expression
r

f (x) dx +
r

z =re , 0t

it

f (z ) dz =

a ie sin a, 2

Ch: 1

187 (Soln-22)

TOC

Index

23 we may let r to obtain

xeiax dx = iea sin a. x4 + 4 2

Take imaginary parts to obtain

x sin ax dx = ea sin a. 4 x +4 2

(b) By the analysis of (a), the integral is 2i times the sum of the residues in the

-r

r
Figure S4.2.3

upper half plane of z (z 2 + 1)(z 2 + 2z + 2) The residue at z = i is 1 1 2i i = = . 2i(i2 + 2i + 2) 2(1 + 2i) 10 The residue at 1 + i is 1 + i 1 + i 1 + 3i 1 = = . [(1 + i)2 + 1] 2i 4 + 2i 10 (c) The integral is 2i residues of 1/(z 2 4z + 5)2 in the upper half plane. Now 2 2 z 4z + 5 = (z 2) + 1, so there are poles of order 2 at 2 + i and 2 i. By (4.2.2d), he residue at 2 + i is 1 d dz (z (2 i))2 The integral is 2i/4i = /2. (d) The integral is
2 0

z . (z i)(z + i)(z (1 + i))(z (1 i))

Thus the integral is 2i(i/10) = /5.

=
z =2+i

2 (z (2 i))3

=
z =2+i

2 . 8i3

ei + ei d = 2(5 + 2(ei + ei ))

|z |=1

dz z + z 1 1 2(5 + 2(z + z )) iz

Ch: 1

188 (Soln-23)

TOC

Index

24 which is 2i times the sum of the residues of the integrand inside the unit circle. Multiply numerator and denominator of the integrand by z to get z2 + 1 z2 + 1 = . 2iz (2z 2 + 5z + 2) 2iz (2z + 1)(z + 2) The residue at z = 0 is 1/4i, and the residue at z = 1/2 is 5 5/4 = . 2i(1/2)2(3/2) 12i Thus the integral is 2i(2/12i) = /3. (e) Since the integrand is an even function, we may integrate from to and divide by 2 to get i residues of 1/(z 4 + a4 ) in the upper half plane. By a computation similar to (a), the residue at aei/4 is 1/(4a3 ei3/4 ), and the residue at aei3/4 is 1/(4a3 ei9/4 ). Thus the integral is i i/4 3 (e + ei3/4 ) = 3 (sin + sin ) = 2 3. 4a3 4a 4 4 4a (f) We may integrate eix /(x2 + 1) from to , divide by 2, and take the real part to get Re(i residues of eiz /(z 2 + 1) in the upper half plane). The residue at z = i is e1 /2i, hence the integral is Re(ie1 /2i) = /2e. (g) The integral is
2 0

ei ei 2i

2n

d =
|z |=1

z z 1 2i

2n

dz iz

which is 2i times the residue of (z 2 1)2n /(i22n z 2n+1 )(1)n at z = 0. But the Taylor expansion of (1 z 2 )2n is 1 2nz 2 + 2n 4 2n 6 2n 2 n z z + + (1)n z + + z 4n . 2 3 n
2n n

Thus the coecient of 1/z in the Laurent expansion of (z 2 1)2n /z 2n+1 is (1)n Therefore
2

(sin )2n d =
0

n 2 2n 2 (2n)! = n 2. 2 n 2 (2 n!)

Remark : In these examples [except for (d) and (g)] we needed a result of the form f (z ) dz 0 as r .

z =reit 0t

By (2.1.5), this will hold if zf (z ) 0 as z in the upper half plane.

Ch: 1

189 (Soln-24)

TOC

Index

25 5. When z = i(2n + 1) , n an integer, 1 + ez = 0. These are simple poles of f (z ) = (Log z )/(1 + ez ) with residues Log(i(2n + 1) )/ei(2n+1) = Log i(2n + 1) . Since n(, i ) = 1, the integral is 2i[Res(f, i3 ) Res(f, i )] = 2i[ Log(i3 ) + Log(i )] = 2i[ ln 3 i ] = 2 2 i2 ln 3. 6. (a) The Taylor expansion of sin2 z has no term of degree 3, so the residue is 0. (b) The Taylor expansion of z 3 sin z 2 is z 5 [1 z4 z8 z 12 + + ] 3! 5! 7!

and by long division, the reciprocal of the expression in brackets has a z 4 term with coecient 1/3! = 1/6. The residue is therefore 1/6. (c) We have z cos 1 1 1 =z 1 + 2 z 2!z 4!z 4

and the residue is therefore -1/2. 7. We have sin( where fn (z ) = [(1)(n1)/2 ] and fn (z ) = 0 for n even. Now for n odd, fn (z ) = (1)(n1)/2 n2 z 2 (1 + nz + + ) = n n!z 2!

ez e3z e5z ez fn (z ), z = 0, )= + = 3 z z 3!z 5!z 5 n=1 enz , n!z n

n odd

akn z k , z = 0,
k=

where akn = 0, k < n, and the series is the Laurent expansion of fn about z = 0. Now

n=1 k=

|akn z k | =

n=1 n odd

en|z| = sinh n!|z |n

e|z| |z |

< .

Thus we may reverse the order of summation to obtain

sin(ez /z ) =
k= n=1

akn

zk ,

z = 0.

Ch: 1

190 (Soln-25)

TOC

Index

26 This is the Laurent expansion of sin(ez /z ) about z = 0. The residue at z = 0 is therefore n=1 a1,n . But a1,n = 0 for n even, and for n odd we have a1,n = (n1)/2 n1 (1) n /(n 1)!n!. Thus the residue is (1)(n1)/2 nn1 . (n 1)!n! n=1,3,5, 8. (a) Since sin lies above the line segment joining (0,0) to (/2, 1), we have sin 2/, 0 /2. Thus
/2 0 /2

er sin d

e2r/ d =
0

(1 er ). 2r

(b) Let be the path of Figure 4.2.3, traversed in the positive sense; consists of a radial path 1 away from z0 , followed by , and completed by a radial path 2 toward z0 . If g (z ) = f (z ) [k/(z z0 )], k = Res(f, z0 ), then g is analytic at z0 , so g (z ) dz = 0 by Cauchys theorem. Now f (z ) dz =

g (z ) dz + k

dz z z0 g (z ) dz + k

g (z ) dz

dz z z0 .

Since the integrals along 1 and 2 approach 0 as 0 by (uniform) continuity of dz g , we must show that z z0 dz i. In fact, if 0 is the angle between 1 and the horizontal, then dz = z z0
0 + 0

iei d = i. ei

9. (a) By Problem 8a, the integral around the large semicircle approaches 0 as the radius approaches . By Problem 8b, the integral around the small semicircle approaches i Res(eiz /z, 0) = i as the radius approaches 0. It follows that (eix /x) dx i = 0 (where the integral is interpreted as a Cauchy principal value), or [(sin x)/x] dx = . (b) By Cauchys theorem,
r

0=
0

eix dx +
0

/4

exp(ir2 ei2t )ireit dt +

r

exp(is2 ei/2 )ei/4 ds.

The second integral is, in absolute value, less than or equal to

/4

r
0

er

sin 2t

dt 0 as r

(sin 2/, 0 /2; see Problem 8 for details). Thus

eix dx =
0 0

es ei/4 ds =

1 i/4 e , 2

Ch: 1

191 (Soln-26)

TOC

Index

27 and therefore
0

cos x2 dx =

sin x2 dx =

1 2

2.

(c) The integral of [Log(z + i)]/(z 2 + 1) on is 2i times the residue at z = i of [Log(z + i)]/(z 2 + 1), which is 2i(Log 2i)/2i = ln 2 + i 2 /2. (Note that Log(z + i) is analytic except for z = i x, x 0.) Thus
0 r

Log(x + i) dx + x2 + 1

r 0

i 2 Log(x + i) dx ln 2 + . x2 + 1 2

(The integral around the large semicircle approaches 0 as r , by the M-L theorem.) Now let x = x in the rst integral to obtain
r 0

[Log(i x) + Log(i + x)] 2 dx ln 2 + i . x2 + 1 2

But Log(i x) + Log(i + x) = ln[|i x||i + x|] + i(1 + 2 ) = ln(x2 + 1) + i (see Figure S4.2.4). Hence
0

ln(x2 + 1) dx + i x2 + 1
0

i 2 dx = ln 2 + x2 + 1 2

or ln(x2 + 1) dx = ln 2. x2 + 1

.i
i + x 2 -x i - x 2 1 x
Figure S4.2.4 (d) In (c) let x = tan to obtain
/2

ln 2 =
0

ln(tan2 + 1) sec2 d = 2 tan2 + 1

/2 0

/2

ln cos d
0

so ln cos d = ln 2. 2

Set =

x to get ln 2 = 2
0

ln cos(
/2

x) dx = 2

/2

ln sin x dx.
0

Ch: 1

192 (Soln-27)

TOC

Index

28 10. Let f (z ) = z 4 , g (z ) = z 4 + 6z + 3. Then |f (z ) g (z )| = |6z + 3|; if |z | = 2, this is less than or equal to 12 + 3 < |z |4 = |f (z )|. Since f has all its zeros inside {z : |z | = 2}, so does g . Now let f (z ) = 6z, g (z ) = z 4 + 6z + 3. Then |f (z ) g (z )| = |z 4 + 3| 4 < 6|z | for |z | = 1. Thus g has one root inside {z : |z | = 1}, hence there are 3 roots in {z : 1 < |z | < 2}. (Since |f g | < |f | when |z | = 1, g cannot be 0 when |z | = 1.)

12. Apply the hexagon lemma (3.4.5) to the compact set K0 = {z : |f (z )+ g (z )| = |f (z )| + m |g (z )|}. If 1 , . . . , m are the polygonal paths given by the lemma, let = j =1 j . Then \ K0 , so |f + g | < |f | + |g | on . Since Z (f ) Z (g ) K0 , we have n(, z ) = 1 for each z Z (f ) Z (g ). Again by (3.4.5), is -homologous to 0. The result now follows from (4.2.9).

11. Apply Rouch es theorem to f (z ) z n and z n . We have |f (z ) z n + z n | = |f (z )| < n | z | when |z | = 1. Since z n has n zeros inside the unit circle, so does f (z ) z n .

13. First let u 0. Then the integral is 2i times the residue of eiuz /[ (1 + z 2 )] at z = i, which is 2ieu /2i = eu . Now let u < 0. Then |eiu(x+iy) | = euy is bounded on {x + iy : y 0} but not on {x + iy : y 0}. Thus we must complete the contour in the lower half plane, as indicated in Figure S4.2.5. Therefore the integral is 2i times the residue of eiuz /[ (1 + z 2 )] at z = i, which is 2ieu / 2i = eu .

y x

Figure S4.2.5 14. We have

esin 1/z =
k=0

[sin(1/z )]k . k!

Since | sin 1/z | is bounded on {z : |z | = 1}, the Weierstrass M -test shows that the series converges uniformly on this set and may therefore be integrated term by term. The residue theorem yields esin 1/z dz = 2i
|z |=1

1 Res(sinn 1/z, 0). n ! n=0

But sin 1 1 1 1 = + , z = 0, 3 z z 3!z 5!z 5

Ch: 1

193 (Soln-28)

TOC

Index

29 and thus all residues of (sin 1/z )n at z = 0 are 0 except for n = 1, in which case the residue is 1. The integral is therefore 2i. 15. (a) Near z0 we have f (z ) (z z0 )k [ak + ak+1 (z z0 ) + ] = . g (z ) (z z0 )k+1 [bk+1 + bk+2 (z z0 ) + ] The residue is therefore ak bk+1 = g (k+1) (z f (k) (z0 ) f (k) (z0 )/k ! = (k + 1) (k+1) . g (z0 ) 0 )/(k + 1)!

(b) Near z0 we have f (z )/g (z ) = [a0 + a1 (z z0 ) + ]/[(z z0 )2 h(z )] where h(z ) = b0 + b1 (z z0 ) + . The residue is therefore h(z0 )f (z0 ) f (z0 )h (z0 ) d [f (z )/h(z )]z=z0 = = [f (z0 )/b0 ] [f (z0 )b1 /b2 0 ]. dz h2 (z0 ) But g (z ) = b0 (z z0 )2 + b1 (z z0 )3 + , so b0 = g (z0 )/2! and b1 = g (z0 )/3!, and the result follows. 16. let f (z ) = 3z, g (z ) = 3z ez . then |f (z ) g (z )| = |ez | = |e(x+iy) | = ex e < 3 = |f (z )| for |z | = 1. The result follows from Rouch es theorem. 17. If w D(0, ), we must show that w = f (z ) for some z D(0, r), that is, f w has a zero in D(0, r). Now when |z | = r we have |(f (z ) w) f (z )| = |w| < |f (z )| by hypothesis. By Rouch es theorem, f w and f have the same number of zeros in D(0, r). But f has at least one zero in D(0, r) since f (0) = 0, and the result follows. 18. The analytic function 1/ez contributes zero to the integral, as does cos 1/z , whose residue at 0 is 0. Since +i is inside the circle C (1 + i, 2) but i is outside, the integral is 2i times the residue of ez /[(z i)(z + i)] at z = i. Thus the integral is 2i(ei /2i) = .

19. Let r be the contour formed by traveling from r to r along the real axis, and then returning to r on the semicircle S (0, r) (in the upper half plane) with center at 0 and radius r. The integral of P (z )/Q(z ) on the semicircle approaches 0 as r , by the M-L theorem. For r suciently large, r encloses all the poles of P/Q in the upper half plane, so P (z ) dz = Q(z )
r r

P (x) dx + Q(x)

S (0,r )

P (z ) dz Q(z )

and we may let r to get the desired result. For the specic example, note that the poles of z 2 /(1 + z 4 ) in the upper half plane are at z = ei/4 and ei3/4 . The residues are
z ei/4

lim

(z ei/4 )z 2 1 ei/2 (z ei/4 ) i/2 = ei/4 = e lim = z4 + 1 z4 + 1 4 4ei3/4 z ei/4

Ch: 1

194 (Soln-29)

TOC

Index

30 and
z ei3/4

lim

ei3/2 (z ei3/4 )z 2 1 = = ei3/4 . z4 + 1 4 4ei9/4

Thus the integral is 2i i/4 1 i 3 i (e + ei3/4 ) = (i sin i sin ) = (i 2) = 2. 4 2 4 4 2 2 20. Apply Rouch es theorem with f (z ) = az n and g (z ) = az n ez . Then for |z | = 1, |f (z ) g (z )| = |ez | e|z| = e < |a| = |f (z )|, and the result follows. 21. Let f (z ) = 2z, g (z ) = 2z + 1 ez . then for |z | = 1, |f (z ) g (z )| = |ez 1| = |z + so |f (z ) g (z )| 1 + 22. 23. 1 1 + + = e 1 < 2 = |f (z )| 2! 3! z2 z3 + + | 2! 3!

24.

25.

26.

and Rouch es theorem applies. Let g (z ) = 5z 4 . If |z | = 1, then |f (z ) g (z )| = |z 7 + z 2 2| 1 + 1 + 2 < |g (z )| and Rouch es theorem applies. If g (z ) = z 5 , then for |z | = 2 we have |f (z ) g (z )| = |15z + 1| 31 < 25 = |g (z )|. If h(z ) = 15z , then for |z | = 1/2, |f (z ) h(z )| = |z 5 + 1| (1/2)5 + 1 < 15/2 = |h(z )|. The result follows from Rouch es theorem. Apply Rouch es theorem with f (z ) = z 5 , g (z ) = z 5 + z + 1. We have, for |z | = 5/4, |f (z ) g (z )| = |z + 1| (5/4) + 1 = 9/4. But |f (z )| = (5/4)5 = 3.05 > 9/4, and the result follows. If fn (zn ) = 0 for all n and zn 0, then |f (z0 )| |f (z0 ) f (zn )| + |f (zn ) fn (zn )| + |fn (zn )| 0 as n by the uniform convergence of fn on compact subsets and the continuity of f at z0 . Thus f (z0 ) = 0. Conversely, assume f (z0 ) = 0. Since f is not identically zero, there is a disk D(z0 , r) containing no zero of f except z0 . Let = min{|f (z )| : |z z0 | = r} > 0. For suciently large m, |f (z ) fm (z )| < es for all z D(z0 , r), hence on C (z0 , r) we have |f (z ) fm (z )| < |f (z )|. By Rouch theorem, fm has a zero in D(z0 , r), say at zm . We may repeat this process using the disks D(z0 , 2n r), n = 1, 2, 3, . . . to nd the desired subsequence. (a) This is a direct calculation. (b) By hypothesis, p must have n k zeros in |z | > 1, and the result follows from (a). (c) This follows from (a) if we note that for |z | = 1, we have zz = 1, hence 1/z = z . (d) Assume |a0 | > |an |. If g (z ) = a0 p(z ), then |f (z ) g (z )| = |an q (z )| < |a0 p(z )| by part (c), so |f (z ) g (z )| < |g (z )|. By Rouch es theorem, f has k zeros in |z | < 1. Now assume |a0 | < |an |. If h(z ) = an q (z ), then for |z | = 1, |f (z ) h(z )| = |a0 p(z )| < | an q (z )| = |h(z )|. By Rouch es theorem and part (b), f has n k zeros in |z | < 1. (e) If |a0 | > |an | and p has no zeros in |z | > 1, then p has n zeros in |z | < 1, hence so does f , by (d). If |a0 | < |an | and p has no zeros in |z | < 1, then by (d), f has n zeros in |z | < 1. In either case there is a contradiction, because f is a polynomial of degree at most n 1.

Ch: 1

195 (Soln-30)

TOC

Index

31

Section 4.3
1. Near z0 we have f (z ) = n=1 an (z z0 )n and g (z ) = m=0 bm (z z0 )m . The Laurent expansion of g (z )f (z ) is found by multiplying the two series, and Res(gf, z0 ) = b0 a1 = g (z0 ) Res(f, z0 ), as desired. For the counterexample, take z0 = 0, f (z ) = (1/z 2 ) + (1/z ), g (z ) = 1 + z . Then Res(gf, 0) = 2; on the other hand, g (0) = 1, Res(f, 0) = 1. 2. If f (z0 ) = w0 , then since f is one-to-one, k = minzC (z0 ,r) |f (z ) w0 | > 0. Thus if |w w0 | < k , we may expand 1 1 1 = = f (z ) w f (z ) w0 (w w0 ) f (z ) w0 1 1
ww0 f (z )w0

in a geometric series. Term by term integration shows that f 1 is analytic at w0 . by 3. Let z0 P . If r is suciently small, then V = {1/f (z ) : z D(z0 , r)} is open in C because the image under 1/z of a disk (4.3.1). Also, W = {1/z : z V } is open in C containing 0 is a neighborhood of . But W = f (D(z0 , r)), and the result follows. 4. By the residue theorem, the integral is j =1 Res(gf /f, aj ). Since Res(gf /f, aj ) = m(f, aj )g (aj ) by (4.2.2e) and Problem 1 of this section, the result follows.
n

5. If z0 and D(z0 , r) , then the image of D(z0 , r) under f will contain a disk D(f (z0 ), s). Since D(f (z0 ), s) will contain points w1 , w2 , w3 such that |w1 | > |f (z0 )|, Re w2 > Re f (z0 ), and Im w3 > Im f (z0 ), it follows that |f |, Re f , and Im f cannot take on a local maximum at z0 .

Sections 4.4 and 4.5

1. For the inverse, solve w = (az + b)/(cz + d) for z . For the composition, consider w = (au + b)/(cu + d), u = (z + )/(z + ) and substitute. Alternatively, use the fact that a linear fractional transformation is a composition of maps of types (i)-(iv) of (4.4.1). 2. (a) If w = (1 + z )/(1 z ) then z = (w 1)/(w + 1), so T 1 (w) = (w 1)/(w + 1). (b) It is easier to deal with T 1 . Figure S 4.5.1 shows that T 1 maps Re w > 0 onto |w| < 1, {Re w = 0} {} onto |w| = 1, and Re w < 0 onto |w| > 1; the result follows.

Im

.
+1 1

.
-1

. 1

Re

Figure S4.5.1

Ch: 1

196 (Soln-31)

TOC

Index

32 3. (a) Possibly motivated by the analysis of Problem 2, try T (z ) = k (z i)/(z + i). Since T (1) = 1 we have k = (1 + i)/(1 i), and this does yield T (1) = 1, as desired. (b) The desired transformation is accomplished by an inversion followed by a 180 degree rotation, in other words, T (z ) = 1/z . 4. (a) T must be of the form T (z ) = k (z z1 )/(z z3 ). Since T (z2 ) = 1 we have 1 = k (z2 z1 )/(z2 z3 ), which determines k uniquely. (b) If z1 = then T (z ) = (z2 z3 )/(z z3 ). If z2 = then T (z ) = (z z1 )/(z z3 ), and if z3 = then T (z ) = (z z1 )/(z2 z1 ). (c) If T1 is the unique linear fractional transformation mapping z1 , z2 , z3 to 0, 1, , and T2 is the unique linear fractional transformation mapping w1 , w2 , w3 to 0, 1, , 1 then T = T2 T1 . (If T is another linear fractional transformation mapping z1 , z2 , z3 to w1 , w2 , w3 , then T2 T maps z1 , z2 , z3 to 0, 1, . Thus T2 T = T1 , hence 1 T1 = T , proving T unique.) T = T2

5. (a) This follows from the fact that f is one-to-one. (b) This is a consequence of the open mapping theorem for meromorphic functions (Section 4.3, Problem 3). by part (b). If is an essential singularity, (c) Let w f (D(0, 1)), which is open in C then by the Casorati-Weierstrass theorem we nd zn with f (zn ) w. Thus for large n, zn / D(0, 1) but f (zn ) f (D(0, 1)), contradicting the assumption that f is one-to-one. , then f is constant by Liouvilles theorem. Thus by part (d) If f is analytic on C (a), there is only one remaining case to consider, in which f has poles at and at \ D(z0 , 1)) are disjoint open sets in C . Since z0 C. As in (b), f (D(z0 , 1)) and f (C \ D(z0 , 1))is a bounded set, that is, f is f (D(z0 , 1)) (because f (z0 ) = ), f (C bounded on the complement of D(z0 , 1). This contradicts the assumption that is a pole. (e) If z0 = , then by Problem 9(c) of Section 4.1, f is a polynomial, and deg f = 1 because f is one-to-one. If z0 C, then since f has a pole at z0 , g is analytic at z0 . By the open mapping theorem (4.3.1), g (z0 ) = 0. (If g (z0 ) = 0 then g , hence f , is not one-to-one. (f) If z0 = , this follows from (e), so assume z0 C. By (e), g (z0 ) = 0, hence (z z0 )f (z ) = (z z0 )/(g (z ) g (z0 )) 1/g (z0 ) as z z0 . By part (b) of (4.1.5), f has a simple pole at z0 . (g) Let h(z ) = f (z ) [Res(f, z0 )/(z z0 )]. By (4.2.2d), limzz0 (z z0 )h(z ) = 0. Thus and is therefore constant. h(z ) has only removable singularities in C

Section 4.6
1. By (4.6.3i), 1 f (a)f (z ) f (z ) f (a) ; za 1 az let z a to obtain (4.6.3ii).

Ch: 1

197 (Soln-32)

TOC

Index

33 2. Since Re z > 0, we have |w f (0)| < |w (f (0))| for Re w > 0 (draw a picture). Thus T maps {w : Re w > 0} into D(0, 1), so T f is an analytic map of D(0, 1) into itself. Since T (f (0)) = 0, Schwarzs lemma implies that |T (f (z )| |z |, z D(0, 1), that is, |f (z ) f (0)| |z ||f (z ) + f (0)|. Thus both |f (z )| |f (0)| and |f (0)| |f (z )| are less than or equal to |z |[|f (z )| + |f (0)|]. This yields the rst statement of the problem. Now f (z ) + f (0) (f (z ) f (0)) d T (f (z )) = f (z ), dz [f (z ) + f (0)]2 and this is at most 1 in absolute value when z = 0, by Schwarzs lemma. Thus |2 Re f (0)| |f (0)| 1 |2 Re f (0)|2 and the result follows. 3. If f (z0 ) = z0 and f (a) = a, with z0 = a, then equality holds at z0 in (4.6.3i). In this 1 1 case b = f (a) = a, so f = a a with || = 1. Now z0 = f (z0 ) = a (a (z0 )), 1 hence a (z0 ) = a (z0 )). Since z0 = a, we have a (z0 ) = 0, so = 1 and f = a a , the identity function. 4. (a) The function f must have the form given in (4.6.6) in D(0, 1), hence on C by the identity theorem. Since f is entire, the only possibility is n = 1, a1 = 0, so f (z ) = z k for some unimodular and nonnegative integer k . (b) Let the poles of f in D(0, 1) be at b1 , . . . , bm , with orders l1 , . . . , lm respectively. Then by (4.6.6), f is of the form f (z ) =
n j =1 m j =1 z aj ) 1aj z z bj ) 1bj z kj lj

with || = 1; aj , bj D(0, 1); kj , lj = 0, 1, . . . . (Note that f (z ) times the denominator of the above fraction has only removable singularities in D(0, 1).) 5. The function g satises the hypothesis of (4.6.3), so by (4.6.3i), g (z ) g (a) za , 1 az a, z D,

1 g (a)g (z ) that is,

M (f (Rz ) f (Ra)) M 2 f (Ra)f (Rz ) Let w = Rz, w0 = Ra, to obtain M (f (w) f (w0 )) M2 f (w0 )f (w)

za . 1 az

R(w w0 ) , R2 w 0 w

w, w0 D(0, R)

Ch: 1

198 (Soln-33)

TOC

Index

34 which is the desired generalization of (i). By (4.6.3ii), |g (a)| (1 |g (a)|2 )/(1 |a|2 ), that is, R 1 [|f (Ra)|2 /M 2 ] |f (Ra)| . M 1 |a|2 Thus |f (w0 )| or (M/R) [|f (w0 )|2 /M R] , 1 |w0 /R|2 R(M 2 |f (w0 )|2 ) M (R2 |w0 |2 )

|f (w0 )| which generalizes (ii). 6. Let g (z ) =

f (z )
n j =1 z zj 1z j z kj

Then g is analytic on D(0, 1), continuous on D(0, 1), and |g (z )| = |f (z )| 1 when |z | = 1. The assertion now follows from the maximum principle. If equality holds at some point z0 in D(0, 1) (other than the zj ), then |g (z0 )| = 1, so g is constant by the maximum principle. Thus
n

f (z ) = c
j =1

z zj 1 zj z

kj

where c is a constant with |c| 1.

Section 4.7
1. If |z | < 1, then (2i)1 (2 )
1 w+z |w|=1 w(wz )

e +z eit z
it

dt = 1, as desired.

dw = 1 + 2 = 1 by the residue theorem. Thus

2. Since 1 cos( t) 1, we have 1r 1 r2 1 r2 1+r = P ( t ) = . r 1+r (1 + r)2 (1 r)2 1r The result now follows from (4.7.8) and the observation that by (4.7.9), u(0) = 1 2
2

u(eit ) dt.
0 2

3. If D(z0 , R) , then by (4.7.8) with r = 0, un (z0 ) = (2 )1 0 un (z0 + Reit ) dt. Let 2 n to obtain u(z0 ) = (2 )1 0 u(z0 + Reit ) dt. By (4.7.10), u is harmonic on .

Ch: 1

199 (Soln-34)

TOC

Index

35 4. It is sucient to consider the case where u is continuous on D(0, 1) and analytic 2 on D(0, 1). Then by (4.7.8), u(z ) = (2 )1 0 Pz (t)u(eit ) dt, |z | < 1. Let f (z ) = 2 (2 )1 0 Qz (t)u(eit ) dt. Then f is analytic on D(0, 1) by (3.3.3), and Re f = u by (4.7.2), as desired. 5. (i) We have [z0 , z, ) = ([z0 , z, ) ) ([z0 , z, ) ) ([z0 , z, ) (C \ )). The rst and third sets on the right are nonempty, relatively open subsets of [z0 , z, ). Since [z0 , z, ) is connected, [z0 , z, ) = . Let be any point in [z0 , z, ) . It follows from (a) and (b) that [z0 , ) . (See Figure 4.7.1 to visualize this.)

Now either z (z0 , ) or (z0 , z ). If (z0 , z ), we can repeat the above argument with z0 replaced by to get 1 such that 1 (, z, ). But then (a) and (b) imply that , a contradiction. Thus z (z0 , ), hence [z0 , z ] [z0 , ) . (ii) We have f (w) dw = 0 by (3.3.1). Since | (t) (t)| = (1 )| (t) z0 | 0 as 1, uniformly in t, it follows from the uniform continuity of f on compact sets that we may let 1 to obtain f (w) dw = 0. The result n(, z )f (z ) = (2i)1 [f (w)/(w z )] dw is obtained similarly. (Note that n( , z ) = n(, z ) for all suciently close to 1, by (3.2.3) and (3.2.5).] 6. The two equations given in the outline follow immediately from Problem 5. Subtract the second equation from the rst to obtain f (z ) = 1 2i f (w)

1 1 wz wz

dw.

If z = x + iy, w = t + i , then 1 1 zz 2iy = = . wz wz (w z )(w z ) [t x + i( y )][t x + i( + y )] 1

R

If w is real, so that = 0, this becomes 2iy/[(t x)2 + y 2 ]. Thus f (z ) =

R

yf (t) 1 dt + (t x)2 + y 2

yf (w) dw (w z )(w z )

where R is the semicircular part of the contour. Let Mf (R) be the maximum value of |f | on R . By the M-L theorem, for large R the integral around R is bounded in absolute value by a constant times Mf (R)/R, so that if Mf (R)/R 0 as R , we obtain f (z ) = lim If |f (z )|/|z |
1

1 R

R R

yf (t) dt. (t x)2 + y 2

0 as z for some > 0, then we may write f (z ) = 1

yf (t) dt (t x)2 + y 2

where the integral exists in the improper Riemann sense, not simply as a Cauchy principal value. Take real parts to get the desired result.

Ch: 1

200 (Soln-35)

TOC

Index

36

Section 4.8
1. We may write f (z ) = cz k + ak+1 z k+1 + ak+2 z k+2 + where c = 0. Apply Jensens formula (4.8.2b) to f (z )/z k to obtain
n( r )

ln |c| = Thus

ln
j =1

1 aj + r 2

ln
0

f (reit ) dt. rk

n( r )

k ln r + ln |c| =

ln
j =1

aj 1 + r 2

2 0

ln |f (reit )| dt.

But c = f (k) (0)/k !, and the result follows. 2. The statement is

n m

ln |f (0)| =

kj ln
j =1

bj aj 1 lj ln + R R 2 j =1

2 0

ln |f (Reit )| dt.

3. First note that if 0 < r < R, then n(t) is a step function on [0, r] which is left continuous, having jumps only at the radii of those circles that pass through zeros of f . To avoid cumbersome notation, we illustrate the ideas with a concrete example Suppose 0 < |a1 | = |a2 | = |a3 | < |a4 | < |a5 | = |a6 | < |a7 | < r |a8 |. Then the graph of n(t), 0 t r, is shown in Figure S4.8.1. Since n(t) is constant between jumps and (1/t) dt = ln t, we have
r 0

To prove the statement, note that we may write f = g/h, where g has zeros at a1 , . . . , an , h has zeros at b1 , . . . , bm , and g and h each satisfy the hypothesis of (4.8.1). Since ln |f | = ln |g | ln |h|, the result follows.

n(t) dt = n(|a3 |) ln |a3 | + n(|a4 |)(ln |a4 | ln |a3 |) t + n(|a6 |)(ln |a6 | ln |a4 |) + n(|a7 |)(ln |a7 | ln |a6 |) + n(r)(ln r ln |a7 |).

If we observe that |a7 | < r |a8 |, so that n(r) n(|a8 |), we may write
r 0

n(t) dt = ln |a3 |[n(|a4 |) n(|a3 |)] t ln |a4 |[n(|a6 |) n(|a4 )] ln |a6 |[n(|a7 |) n(|a6 |)] ln |a7 |[n(|a8 |) n(|a7 |)] + n(r) ln r.

Now ln |a6 |[n(|a7 |) n(|a6 |)] = 2 ln |a6 | = ln |a5 | ln |a6 |

Ch: 1

201 (Soln-36)

TOC

Index

37

n(t)

7 6

o o

. .
t-axis |a |
7

4 3 o

.
|a4 | |a 6| r

.
|a 3|
Figure S4.8.1

and similarly for the other terms. Thus

r 0

n(t) dt = ln |aj | + n(r) ln r = t j =1

n( r )

n( r )

ln
j =1

r |aj |

as desired. 4. By Problem 3,
r 0

n(t) dt = t

n( r )

ln
j =1

r . |aj |

Also, ln[|f (k) (0)|rk /k !] = k ln r + ln[|f (k) (0)|/k !]. The result now follows from (4.8.5) if we observe that 1 2
2 0

ln |f (reit | dt

1 2

ln M (r) dt = ln M (r).
0

5. By (4.8.5) and Problem 3, 1 2

2 0 r

ln |f (reit | dt = k ln r + ln[|f (k) (0)/k !] +

n(t) dt, t

which is a continuous, increasing function of r. Each time n(t) has a jump, say a jump r of size c at t = r0 (see Figure S4.8.1), 0 [n(t)/t] dt contributes a term of the form c(ln r ln r0 ), r > r0 .

Ch: 1

202 (Soln-37)

TOC

Index

38

Section 4.9
1. If z = rei2p/q , then z n! = rn! ei2n!p/q = rn! if n q , and rn! 1 as r 1. It follows that any analytic function that agrees with f on D(0, 1) cannot approach a nite limit as z approaches a point on C (0, 1) of the form ei2p/q . Since these points are dense in C (0, 1), there can be no extension of f to a function analytic on D(0, 1) D(w, ), |w| = 1. 2. (a) Let S = {x + iy : y = 0, x 0}. If z1 / S , let D1 , . . . , Dn be disks such that Di Di+1 = , i = 1, . . . , n 1, 1 D1 , z1 Dn , and Di S = , i = 1, . . . , n. Let fi (z ) = Log z, z Di . Then (fn , Dn ) is a continuation of (f1 , D1 ) = (f, D) relative to , so fn , Dn ) . If, say, z1 is in the second quadrant, then Log z1 = ln |z1 | + i(z1 ) where can be be chosen in the interval [0, 2 ). If z2 S, z2 = 0, let E1 , . . . , Em be disks such that E1 = Dn (so z1 E1 ), Ei Ei+1 = , i = 1, . . . , m 1, z2 Em , and Ei T = , i = 1, . . . , m, where T = {x + iy : y = 0, x 0}. Let gi (z ) = log z = ln |z | + i(z ), 0 < 2, z Ei . Then (gm , Em ) is a continuation of (g1 , E1 ) = fn , Dn ) relative to , so (gm , Em ) . (b) By the argument of (a), if there were such an h, then h(z ) = Log z, z / S , and hence h must be discontinuous on the negative real axis, a contradiction. 3. The reasoning beginning with since power series converge absolutely is faulty. If k / D(z0 , r), this does not imply k=0 bk (z z1 ) converges absolutely at some point z that the original series converges at z . For

k=0

|bk | |z z1 | =

k=0 n=k

n an (z1 z0 )nk |z z1 |k < k

does not imply that

k=0 n=k

|an | |z1 z0 |nk |z z1 |k < ,

and the latter is what is needed to reverse the order of summation. 4. If g1 , . . . , gk are analytic on , so is h(z ) = F (z, g1 (z ), . . . , gk (z )), z . (The derivative of h may be calculated explicitly by the chain rule.) It follows that if hj (z ) = F (z, f1j (z ), . . . , fkj (z )), then hj is analytic on Dj , j = 1, . . . , n. Thus (h1 , D1 ), . . . , (hn , Dn ) forms a continuation. But D1 = D and h1 = 0 on D, by hypothesis. By successive application of the identity theorem (2.4.8), we have hn = 0 on Dn , as desired. 5. If (fi+1 , Di+1 ) is a direct continuation of (fi , Di ), then fi = fi+1 on Di Di+1 , hence fi = fi+1 on Di Di+1 . Therefore (fi+1 , Di+1 ) is a direct continuation of (fi , Di ), and the result follows.

Chapter 5
Section 5.1
1+|z | 1. Since |f (z )| 1 |z | |f (0)|, F is bounded, hence F is closed and bounded, and therefore compact. thus F is relatively compact. To show that F is not compact, let fn (z ) =

Ch: 1

203 (Soln-38)

TOC

Index

39 f (z ) 0. By Section 4.5, Problem 2, fn F ; since fn f uniformly on compact subsets of D(0, 1) but f / F , F is not closed, and therefore not compact.
1 1+z n 1z ,

2. We may take a = 0 (if not, consider f a). Since 1/|f (z )| 1/r for all f F and z , by (5.1.10) we have a subsequence {fnk } such that 1/fnk g A(), uniformly on compact subsets. If g is not identically 0, then g is never 0 by (5.1.4), and it follows that fnk 1/g uniformly on compact subsets. If g 0, then fnk uniformly on compact subsets. 3. (a) If F is relatively compact then F is compact, so if fn F , n = 1, 2, . . . , there is a subsequence {fnk } converging to a limit in F (not necessarily in F ). Conversely, if each sequence in F has a convergent subsequence, the same is true for F . (If fn F , choose gn F with d(fn , gn ) < 1/n; if the subsequence {gnk } converges, so does {fnk }). Thus F is compact. (b) F is bounded i F is bounded (by denition of boundedness), i F is closed and bounded (since F is always closed), i F is compact (by the rst statement of (5.1.11)), i F is relatively compact.

n 4. Let F be relatively compact. If f F and f (z ) = n=0 an z , then by (2.4.1), n |an | r max{|f (z )| : |z | = r}, 0 < r < 1. But by compactness, max{|f (z )| : |z | = r} is bounded by a constant M (r) independent of the particular f F . Thus

Mn = sup{|an (f )| : f F} M (r)/rn . Consequently, Mn z n converges if |z | < r, so by (2.2.7), (lim supn Mn )1 r. 1/n Let r 1 to obtain lim supn Mn 1. Conversely, if the desired Mn exist, then if f F and |z | r < 1, we have |f (z )| |an ||z |n n=0 Mn rn < . Thus F is bounded, hence relatively compact. 5. (a) Apply Cauchys formula for a circle to the function f 2 to get, for 0 r < R, f 2 (a) = (the mean value of f 2 ). Thus |f (a)|2 = |f 2 (a)| 1 2
2 0 1/n

1 2

f 2 (a + reit ) dt
0

|f (a + reit )|2 dt.

Now multiply on both sides by r and integrate with respect to r from 0 to R to obtain R2 1 |f (a)|2 2 2
R 2

r
0 0

|f (a + reit )|2 dt dr

6. Let f H (f ) be the suggested map. Since |f | 1 on and f = 0 on the boundary of K , the integral over K is greater than 0 and H is well dened. If fn F and fn f , that is, d(fn , f ) 0, then fn f uniformly on K , hence H (fn ) H (f ), so that H

and the result follows. (b) By part (a), F is bounded, and the result follows from (5.1.10).

Ch: 1

204 (Soln-39)

TOC

Index

40 is continuous. If F were compact, then H (F ) would be a compact, hence bounded, subset of the reals. If 0 < r < R, let f be a continuous function from to [0, 1] such that f = 1 on D = D(a, r) and f = 0 o K (Urysohns lemma). Then |f (x + iy )| dx dy 1 dx dy 1 dx dy
K

7. If z is a point on the open radial line S from 0 to ei , then ei + (1/n)(z ei ) = (1 1/n)ei + (1/n)z also lies on S , and approaches ei as n . By hypothesis, fn converges pointwise on S . Since S certainly has a limit point in S (, ), Vitalis theorem implies that fn converges uniformly on compact subsets. Given > 0 there exists > 0 such that if z S (, ) and |z ei | < , then |z w| < for some w S . It follows that by choosing z suciently close to ei , we can make f (z ) as close as we wish to L, as desired. 8. If k is a complex number, then k will also be used to denote the function that is identically k . Since L(1) = L(12 ) = L(1)L(1), L(1) must be 0 or 1. But if L(1) = 0, then for any f A(), L(f ) = L(f 1) = L(f )L(1) = 0, hence L 0, a contradiction. Thus L(1) = 1, so L(k ) = L(k 1) = kL(1) = k . Now let z0 = L(I ). If z0 / , then h(z ) = 1/(z z0 ) gives h A(). Thus h(I z0 ) = 1, hence L(h)(z0 z0 ) = 1, a contradiction. Therefore z0 . If f A() and g is as dened in the outline, then g A() and g (I z0 ) = f f (z0 ). It follows that L(f ) f (z0 ) = L(g )(L(I ) z0 ) = L(g )(z0 z0 ) = 0.

as r R. Thus H (F ) is unbounded, a contradiction.

9. Dene An as suggested. Then each An is a closed subset of , and since for each z , fk (z ) converges to a nite limit as k , we have n=1 An = . By the Baire category theorem, some An contains a disk D. The fk are uniformly bounded on D, hence by Vitalis theorem, fn f uniformly on compact subsets of D. (Note that D is connected, although need not be.) Thus f is analytic on D. Finally, let U be the union of all disks D such that fn f uniformly on compact subsets of D. Then U is an open subset of and fn f uniformly on any compact K U (because K is covered by nitely many disks). If W is an open subset of , the rst part of the proof shows that W contains one of the disks D whose union is U . Thus U is dense in .

Section 5.2
1. For j = 1, 2, let gj be the unique analytic map of j onto D such that gj (zj ) = 0 and 1 (zj ) > 0 (5.2.3d). Then f = g2 g1 satises f (z1 ) = z2 and f (z1 ) > 0. If h is gj another such map, then g2 h = g1 by (5.2.3d), so h = f . 2. From the denition, h is a continuous map of C into D(0, 1). To prove that h is one-to-one and onto, note that h(rei ) = rei /(1 + r). If h(zn ) h(z ), then h(z ) D(0, r/(1 + r)) for r suciently close to 1. But since h maps D(0, r) one-to-one onto D(0, r/(1 + r)), it follows by compactness that h is a homeomorphism of these sets. Thus zn z , so h1 is continuous.

Ch: 1

205 (Soln-40)

TOC

Index

41 3. By convexity, H (t, s) for all t [a, b] and all s [0, 1]. Since H (t, 0) = (t) and H (t, 1) = (a), the result follows. 4. Proceed as in Problem 3, with the initial point (a) replaced by the star center, to obtain an -homotopy of the given curve to a point (namely the star center). \ Kn . By denition of Kn , we have 5. Let n = C n = {} {z : |z | > n} D(w, 1/n).
wC\

Now consider any component T of n . Since T is a maximal connected subset of n , it follows that T {} {z : |z | > n} or T D(w, 1/n) for some w C \ . In either \ . Since n C \ , T must contain any component of C \ that case, T meets C it meets, and such a component exists by the preceding sentence. 6. (a) Form the sets Kn as in (5.1.1), and nd by (5.2.8) a rational function Rn with poles in S such that |f Rn | < 1/n on Kn . For any compact subset K of , K Kn for suciently large n, so that Rn f uniformly on compact subsets of . \ Kn contains a component of C \ , so if (b) By Problem 5, each component of C is simply connected, i.e., C \ is connected, then C \ Kn is connected for all n. Therefore in part (a), the Rn can be taken to be polynomials. Conversely, assume that for every f A() there is a sequence of polynomials Pn converging to f uniformly on compact subsets of . If is a closed path in , then Pn (z ) dz = 0 for all n, hence f (z ) dz = 0 because is compact. Thus is simply connected. 7. (a) By Runges theorem (see part (b) of Problem 6) there are polynomials pn such that |pn (z ) fn (z )| < 1/n for all z Kn Ln Mn . Then pn 0 pointwise. But if K is any compact set containing all the Bn , then pn cannot approach 0 uniformly on K 1 1. because sup{|pn (z )| : z Bn } 1 n (b) Choose polynomials pn such that |pn (z ) gn (z )| < 1/n for all z Kn Mn . Then pn g pointwise, where g (z ) = 1 for Re z > 0 and g (z ) = 0 for Re z 0.

Section 5.3
1. Let f be a homeomorphism of onto D such that f is a one-to-one analytic map of onto D; f exists by (5.3.9) and (5.2.2). If g = f 1 and u = u0 (g |D ), then u is real-valued and continuous on D, so by (4.7.6), u extends to a function that is continuous on D and harmonic on D. Let u = u f ; then u = u0 on and u is continuous on . If h = u + iv is analytic on D, then h f is analytic on and Re h f = u f = u, hence u is harmonic on . 2. (a) Let u be the unique argument of z in [, ); see (3.1.2). (b) Apply (5.2.2) and (5.3.9). (c) Note that u(f (z )) = Im log (f (z )), and log f (z ) is analytic on D by (3.1.2). (d) Suppose u(f (z ))+iV (z ) is analytic on D. Write V (z ) = v (f (z )) where v is harmonic on . Then iu(f (z ))v (f (z )) is analytic on D, so by (3.1.6), ln |f (z )| = v (f (z ))+2ik for some integer k . Consequently, ev(f (z)) = |f (z )|. If V is bounded, so is v , which yields a contradiction. (Examine f (z ) near z0 , where f (z0 ) = 0.) 3. Apply (5.3.9), along with Problems 3.2.6 and 3.2.7.

Ch: 1

206 (Soln-41)

TOC

Index

42

Chapter 6
Section 6.1
1. If f (z ) = 0, then since fn (z ) 1 as n , it follows that for suciently large N , N 1 the innite product n=N fn (z ) converges. Thus f (z ) = [ k=1 fk (z )]g (z ) where g is N 1 analytic at z and g (z ) = 0. Hence m(f, z ) = k=1 m(fk , z ) = n=1 m(fn , z ). 2. The rst statement is immediate from the power series expansion of ln(1 x), namely x+ x2 x3 1 x x2 + + = x + x2 ( + + + ). 2 3 2 3 4

Now if n an converges, then ln(1 an ) = n [an + g (an )a2 n ] where g (an ) 1/2 as n . By (6.1.1), n (1 an ) converges to a nonzero limit i n a2 n < . The remaining statement of the problem follows similarly. 3. (a) Absolutely convergent by (6.1.2). (b) Does not converge to a nonzero limit by Problem 2, since 1 = . In fact, n (n + 1)
n n (n

+ 1)2 < ,

k=1

(1

1 1 2 n 1 ) = = 0. k+1 2 3 n+1 n+1

(c) Does not converge to a nonzero limit by Problem 2. Here, an = (1)n+1 / n, hence 2 n an = . n an converges but (d) Absolutely convergent by (6.1.2). 4. (a) See Problem 3(c). (b) Take a2n1 = 1/ n and a2n = (1/ n) + (1/n). Remark : This is also an example of an innite product that is convergent but not absolutely convergent. 5. (a) Since n=1 |an z | converges uniformly on compact subsets, the result follows from (6.1.7). (b) Restrict z to a compact set K . For suciently large n (positive or negative), Log (1 z z/n z )e = Log (1 ) + Log ez/n n n (z/n)2 (z/n)3 = + + 2 3 z2 = 2 g (z/n) n

where g (w) 1/2 as w 0. Since K is bounded, there is a constant M such that Log (1 z z/n )e n M n2

for all z K . Thus n Log[(1 z/n)ez/n ] converges uniformly on K . As in the proof of (6.1.6), the innite product converges uniformly on K , so that the resulting function

Ch: 1

207 (Soln-42)

TOC

Index

43 is entire. 1 (c) Since n=2 n(ln n)2 converges, sets and (6.1.6) applies.

|z | n=2 n(ln n)2

converges uniformly on compact sub-

6. If we try to prove that the convergence of zn implies the convergence of zn g (zn ), m m we run into diculty. We would like to argue that | k=n zk g (zk )| k=n |zk g (zk )| 0 as n, m , but this requires the absolute convergence of zn . A similar diculty occurs in the converse direction. [Note that n (1 + zn ) converges absolutely i n zn converges absolutely, by (6.1.2).]

Section 6.2
1. (a) We have m = 0, and the canonical product is (b) The canonical product is greater than (1/b) 1.
n=1 n=1 (1

z/2n ).

Em (z/zn ) where m is the least integer strictly

n=1 [1

(c) We have m = 0, and the canonical product is

2. We may proceed exactly as in (6.2.5), using (6.2.6) in place of (6.2.3).

z/n(ln n)2 ].

Section 6.3
1. By (6.3.7), the result holds for n = 2. For if d is a gcd of {f1 , f2 }, then f1 /d and f2 /d are relatively prime. If (f1 g1 /d) + (f2 g2 /d) = 1, then f1 g1 + f2 g2 = d. To go from n 1 to n, let d be a gcd for {f1 , . . . , fn } and d1 a gcd for {f1 , . . . , fn1 }. Then d is a gcd for {d1 , fn } (by denition of gcd). By the induction hypothesis, we have g1 , . . . , gn1 A() such that f1 g1 + + fn1 gn1 = d1 , and by (6.3.7) there exist h, gn A() such that d1 h + fn gn = d. But then f1 g1 h + + fn1 gn1 h + fn gn = d.

2. Let {an } be a sequence of points in with no limit point in . By (6.2.6) or (6.2.3), there exists fn A() such that Z (fn ) = {an , an+1 , . . . } and m(fn , aj ) = 0, j n. Let I be the ideal generated by f1 , f2 , . . . , that is, I is the set of all nite linear combinations of the form gi1 fi1 + + gik fik , k = 1, 2, . . . , gij A(). If I were principal, it would be generated by a single f . But then Z (f ) Z (h) for each h I , in particular, Z (f ) Z (fn ) for all n. It follows that f has no zeros, so 1 = f (1/f ) I . By denition of I , 1 = g1 f1 + + gn fn for some positive integer n and g1 , . . . , gn A(). Since f1 (a1 ) = f2 (an ) = = fn (an ) = 0, we reach a contradiction.

Ch: 1

208 (Soln-43)

TOC

Index

List of Symbols
C complex plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1 Re real part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1 Im imaginary part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1 arg argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-1 z complex conjugate of z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2 D(a, r) open disk with center at a and radius r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3 D(a, r) closed disk with center at a and radius r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-3 C (a, r) circle with center at a and radius r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-3 f derivative of f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-4 exp exponential function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1-8 (t) dt integral of a complex-valued function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-1 path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2 range of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2 f ( z ) dz path integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-2 Z (f ) zero set of f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-22 log logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-1 n(, z0 ) index (winding number) of z0 with respect to . . . . . . . . . . . . . . . . . . . . . . . . . 3-5 m k cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-11 i=1 i i D (z0 , r) punctured disk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-1 A(z0 , s1 , s2 ) annulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2 Res (f, z0 ) residue of f at z0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7 Pz (t) Poisson kernel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-25 Qz (t) Cauchy kernel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-25 A() analytic functions on . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1 C () continuous functions on . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1 |f |K supremum of |f | on K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-1 innite product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-1 n=1 zn Em (z ) Weierstrass primary factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-6 (x) number of primes less than or equal to x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-1 (z ) Riemann zeta function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-2 von Mangoldt function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-5 a number theoretic function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-5
b a

Ch: 1

209 (List of Symbols-1)

TOC

Index

Index
absolute convergence of an innite product, 6-2 absolute value, 1-1 analytic continuation, 4-41 analytic continuation along a curve, 4-41 analytic function, 1-4 analytic k -th root, 3-8 analytic logarithm, 3-4 analytic mappings of one disk to another, 4-21. angle-preserving property, 4-19 annulus, 4-1 argument, 1-1, 3-1 argument principle, 4-9 argument principle for meromorphic functions, 4-10 Bezout domain, 6-13 big Picard theorem, 4-5 bounded family of functions, 5-3 Casorati-Weierstrass theorem, 4-4 Cauchys estimate, 2-21 Cauchys integral formula, 3-9, 3-11 Cauchys integral formula for a circle, 2-12 Cauchys theorem, 3-9, 3-18 (homology version), 5-11 (homotopic version) Cauchys theorem for starllike regions, 2-6, 2-9 Cauchys theorem for triangles, 2-5, 2-8 Cauchy kernel, 4-25 Cauchy-Riemann equations, 1-6 closed curve or path, 2-2 compactness criterion, 5-5 complex-dierentiability, 1-4 conformal equivalence, 5-10 conformal map, 4-20 conjugate, 1-2 continuous argument, 3-2 continuous logarithm, 3-2 convergence of an innite product, 6-1 convex set, 1-3 cosine function, 2-20 curve, 2-2 cycle, 3-11 derivative, 1-4 direct analytic continuation, 4-41 Dirichlet problem, 4-27, 4-28, 5-28 distance, 1-2 dog-walking theorem, 3-7, 3-8 Enestroms theorem, 1-9

Ch: 1

210 (Index-1)

TOC

Index

2 equicontinuous family of functions, 5-3 equivalent function elements, 4-42 essential singularity, 4-3 Eulers product formula, 7-2 expanding conformal maps to the boundary, 5-18. exponential function, 1-8, 2-19 extended complex plane, 3-13 nitely generated ideal, 6-13 function element, 4-41 fundamental theorem for integrals on paths, 2-3 fundamental theorem of algebra, 2-21 generalized analytic function, 4-42 greatest common divisor, 6-12 harmonic conjugate, 1-9, 5-12 harmonic function, 1-8 Harnacks inequality, 4-30 hexagon lemma, 3-16 holomorphic function, 1-4 homologous curves and cycles, 3-15, 5-11 homotopic curves, 4-42 homotopy, 4-42, 5-11 Hurwitzs theorem, 5-2 hyperbolic functions, 2-20 ideal, 6-13 identity theorem, 2-23 identity theorem for harmonic functions, 2-25 index, 3-5 innite products, 6-1. integral, 2-1, 2-2 integral of the Cauchy type, 2-13 isolated singularity, 4-1 isolated singularity at innity, 4-5 Jensens formula, 4-33, 4-36 Laplaces equation, 1-8 Laurent expansion, 4-3 Laurent series, 4-2 law of permanence of functional equations, 4-45 length of a path, 2-2 LHospitals rule, 2-26 linear fractional transformation, 4-17, 4-18 Liouvilles theorem, 2-21 logarithm, 3-1 logarithmic derivative, 3-4 magnitude, 1-1 M-L theorem, 2-3 maximum and minimum principles for harmonic functions, 2-25

Ch: 1

211 (Index-2)

TOC

Index

3 maximum principle, 2-23, 2-24 meromorphic function, 4-6, 6-9 minimum principle, 2-24 Mittag-Leers theorem, 6-10 M obius transformations, 4-17 modulus, 1-1 monodromy theorem, 4-43 Montels theorem, 5-5 Moreras theorem, 2-14 Noetherian ring, 6-14 open mapping theorem, 4-15 parallelogram law, 1-9 partial fraction expansion, 4-6 path, 2-2 path integral, 2-2 Poisson integral formula, 4-26, 4-27 Poisson integral formula for harmonic functions, 4-29 Poisson kernel, 4-25 Poisson-Jensen formula, 4-32 polarization, 2-1 pole, 4-3 polygonally connected, 1-3 power series, 2-11 prime number theorem, 7-1. primitive, 2-3 principal branch, 3-2 principal ideal, 6-13 principal ideal domain, 6-13 punctured disk, 4-1 ratio test, 2-10 real-dierentiability, 1-6 region, 1-3 relatively compact, 5-5 relatively prime, 6-12 removable singularity, 4-3 residue, 4-7 residue theorem, 4-8 Riemann hypothesis, 7-5 Riemann integral, 2-1 Riemann mapping theorem, 5-8. Riemann sphere, 3-13 Riemann zeta function, 7-2 root test, 2-10 Rouch es theorem, 4-10 Runges theorem, 5-13, 5-17 Schwarzs lemma, 2-26

Ch: 1

212 (Index-3)

TOC

Index

4 Schwarz reection principle, 2-15 second Cauchy theorem, 3-18 separated sets, 1-3 series, 2-10 simple boundary point, 5-20 simple pole, 4-3 simply connected (homologically), 3-19, 5-12 simply connected (homotopically), 4-44, 5-12 sine function, 2-20 singularity, 4-1 star center, 1-3 starlike, 1-3 Tauberian theorem, 7-10 triangle inequality, 1-2 trigonometric functions, 2-20 unimodular, 1-1 unit, 6-12 Vitalis theorem, 5-6 von Mangoldt function, 7-5 Weierstrass factorization theorem, 6-7 Weierstrass M-test, 2-11 Weierstrass products, 6-5 winding number, 3-5 zero set, 2-22

Ch: 1

213 (Index-4)

TOC

Index

Complex Variables, by Ash and Novinger, is available from Prof. Ashs homepage: www.math.uiuc.edu/r-ash/ .

Hyperlinks to the Table of Contents and Index now follow (as prepared by Prof. Girardi).

Ch: 1

214 (hyperlinks)

TOC

Index

Table of Contents
Chapter 1: Introduction 1.1 Basic Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2 Further Topology of the Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.3 Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.4 Real-Dierentiability and the Cauchy-Riemann Equations . . . . . . . . . . . . . . . . . . . . . . . . 9 1.5 The Exponential Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.6 Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Chapter 2: The Elementary Theory 2.1 Integration on Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.2 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.3 The Exponential Function and the Complex Trigonometric Functions . . . . . . . . . . . 33 2.4 Further Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

Chapter 3: The General Cauchy Theorem 3.1 Logarithms and Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.2 The Index of a Point with Respect to a Closed Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 3.3 Cauchys Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 3.4 Another Version of Cauchys Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

Chapter 4: Applications of the Cauchy Theory 4.1 Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 4.2 Residue Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4.3 The Open Mapping Theorem for Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.4 Linear Fractional Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.5 Conformal Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 4.6 Analytic Maps of One Disk to Another . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 4.7 The Poisson Integral Formula and its Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.8 The Jensen and Poisson-Jensen Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 4.9 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

Ch: 1

215 (hyperlinks)

TOC

Index

Chapter 5: Families of Analytic Functions 5.1 The Spaces A() and C() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 5.2 The Riemann Mapping Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 5.3 Extending Conformal Maps to the Boundary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

Chapter 6: Factorization of Analytic Functions 6.1 Innite Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 6.2 Weierstrass Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 6.3 Mittag-Leers Theorem and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

Chapter 7: The Prime Number Theorem 7.1 The Riemann Zeta Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 7.2 An Equivalent Version of the Prime Number Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 157 7.3 Proof of the Prime Number Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

End Matter: Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

Ch: 1

216 (hyperlinks)

TOC

Index

Index. entry: corresponding page within this document (page by authors numbering)
absolute convergence of an innite product : 138 (6-2) absolute value : 5 (1-1) analytic continuation : 102 (4-41) analytic continuation along a curve : 102 (4-41) analytic function : 8 (1-4) analytic k-th root : 50 (3-8) analytic logarithm : 46 (3-4) analytic mappings of one disk to another : 82 (4-21.) angle-preserving property : 80 (4-19) annulus : 62 (4-1) argument : 5 (1-1), 43 (3-1) argument principle : 70 (4-9) argument principle for meromorphic functions : 71 (4-10) Bezout domain : 149 (6-13) big Pichard theorem : 66 (4-5) bounded family of functions : 109 (5-3) Casorati-Weierstrass theorem : 65 (4-4) Cauchys estimate : 35 (2-21) Cauchys integral formula : 51 (3-9), 53 (3-11) Cauchys integral formula for a circle : 26 (2-12) Cauchys theorem : 51 (3-9), 60 (3-18 (homology version)), 117 (5-11 (homotopic version)) Cauchys theorem for starlike regions : 20 (2-6), 23 (2-9) Cauchys theorem for triangles : 19 (2-5), 22 (2-8) Cauchys kernel : 86 (4-25) Cauchy-Riemann equations : 10 (1-6) closed curve or path : 16 (2-2) compactness criterion : 111 (5-5) complex-dierentiability : 8 (1-4) conformal equivalence : 116 (5-10) conformal map : 81 (4-20) conjugate : 6 (1-2) continuous argument : 44 (3-2) continuous logarithm : 44 (3-2) convergence of an innite product : 137 (6-1) convex set : 7 (1-3) cosine function : 34 (2-20) curve : 16 (2-2) cycle : 53 (3-11) derivative : 8 (1-4) direct analytic continuation : 102 (4-41) Ch: 1 2 3 4 5 6 7 217 (hyperlinks) TOC Index

Dirichlet problem : 88 (4-27), 89 (4-28), 134 (5-28) distance : 6 (1-2) dog-walking theorem : 49 (3-7), 50 (3-8) Enestroms theorem : 13 (1-9) equicontinuous family of functions : 109 (5-3) equivalent function elements : 103 (4-42) essential singularity : 64 (4-3) Eulers product formula : 152 (7-2) expanding conformal maps to the boundary : 124 (5-18.) exponential function : 12 (1-8), 33 (2-19) extended complex plane : 55 (3-13) nitely generated ideal : 149 (6-13) function element : 102 (4-41) fundamental theorem for integrals on paths : 17 (2-3) fundamental theorem of algebra : 35 (2-21) generalized analytic function : 103 (4-42) greatest common divisor : 148 (6-12) harmonic conjugate : 13 (1-9), 118 (5-12) harmonic function : 12 (1-8) Harnacks inequaltiy : 91 (4-30) hexagon lemma : 58 (3-16) holomorphic function : 8 (1-4) homologous curves and cycles : 57 (3-15), 117 (5-11) homotopic curves : 103 (4-42) homotopy : 103 (4-42), 117 (5-11) Hurwitzs theorem : 108 (5-2) hyperbolic functions : 34 (2-20) ideal : 149 (6-13) identity theorem : 37 (2-23) identity theorem for harmonic functions : 39 (2-25) index : 47 (3-5) innte products : 137 (6-1.) integral : 15 (2-1), 16 (2-2) integral of the Cauchy type : 55 (2-13) isolated singularity : 62 (4-1) isolated singularity at innity : 66 (4-5) Jensens formula : 94 (4-33), 97 (4-36) Laplaces equality : 12 (1-8) Laurent expansion : 64 (4-3) Laurent series : 63 (4-2) law of permanence of functional equations : 106 (4-45) length of a path : 16 (2-2) Ch: 1 2 3 4 5 6 7 218 (hyperlinks) TOC Index

LHopitals rule : 40 (2-26) linear fractional transformation : 78 (4-17), 79 (4-18) Liouvilles theorem : 35 (2-21) logarithm : 43 (3-1) logarithmic derivative : 46 (3-4) magnitude : 5 (1-1) M-L theorem : 17 (2-3) maximum and minimum principles for harmonic functions : 39 (2-25) maximum principle : 37 (2-23), 38 (2-24) meromorphic function : 67 (4-6), 145 (6-9) minimum principle : 38 (2-24) Mittag-Leers theorem : 146 (6-10) M obius transformations : 78 (4-17) modulus : 5 (1-1) monodromy theorem : 104 (4-43) Montels theorem : 111 (5-5) Moreras theorem : 28 (2-14) Noetherian ring : 150 (6-14) open mapping theorem : 75 (4-15) parallelogram law : 13 (1-9) partial fraction expansion : 67 (4-6) path : 16 (2-2) path integral : 16 (2-2) Poisson integral formula : 87 (4-26), 88 (4-27) Poisson integral formula for harmonic functions : 90 (4-29) Poisson kernel : 86 (4-25) Poisson-Jensen formula : 93 (4-32) polarization : 15 (2-1) pole : 64 (4-3) polygonally connected : 7 (1-3) power series : 25 (2-11) prime number theorem : 151 (7-1.) primitive : 27 (2-3) principal branch : 44 (3-2) principal ideal : 149 (6-13) principal ideal domain : 149 (6-13) punctured disk : 62 (4-1) ratio test : 24 (2-10) real-dierentiability : 10 (1-6) region : 7 (1-3) relatively compact : 111 (5-5) relatively prime : 148 (6-12) Ch: 1 2 3 4 5 6 7 219 (hyperlinks) TOC Index

removable singularity : 64 (4-3) residue : 68 (4-7) residue theorem : 69 (4-8) Riemann hypothesis : 155 (7-5) Riemann integral : 15 (2-1) Riemann mapping theorem : 114 (5-8.) Riemann sphere : 55 (3-13) Riemann zeta function : 152 (7-2) root test : 24 (2-10) Rouch es theorem : 71 (4-10) Runges theorem : 119 (5-13), 123 (5-17) Schwarzs lemma : 40 (2-26) Schwarz reection principle : 29 (2-15) second Cauchy theorem : 60 (3-18) separated sets : 7 (1-3) series : 24 (2-10) simple boundary points : 126 (5-20) simple pole : 64 (4-3) simply connected (homologically) : 61 (3-19), 118 (5-12) simply connected (homotopically) : 105 (4-44), 118 (5-12) sine functions : 34 (2-20) singularity : 62 (4-1) star center : 7 (1-3) starlike : 7 (1-3) Tauberian theorem : 160 (7-10) triangle inequaltiy : 6 (1-2) trigonometric functions : 34 (2-20) unimodular : 5 (1-1) unit : 148 (6-12) Vitalis theorem : 112 (5-6) von Mangoldt function : 155 (7-5) Weierstrass factorization theorem : 143 (6-7) Weierstrass M-test : 25 (2-11) Weierstrass products : 141 (6-5) winding number : 47 (3-5) zero set : 36 (2-22)

Ch: 1

220 (hyperlinks)

TOC

Index